Cloud Hosting

Page 1,742«..10 20..1,7411,7421,743 1,744..1,750 1,760..»

Expert Insights: How to Protect Sensitive Machine-Learning Training Data Without Borking It – DARKReading

October 10, 2022 | Author: admin

Previous columns in this series introduced the problem of data protection in machine learning (ML), emphasizing the real challenge that operational query data pose. That is, when you use an ML system, you most likely face more data-exposure risk than when you train one up in the first place.

In my rough estimation, data accounts for at least 60% of known machine-learning security risks identified by the Berryville Institute of Machine Learning (BIML). That chunk of risk (the 60%) further divides about nine to one with operational data-exposure versus training data-exposure. Training data components account for a minority of data risk in ML, but are an important minority. The upshot is that we need to spend some real energy mitigating the operational data-risk problem posed by ML that we previously discussed, and we also need to consider training data exposure.

Interestingly, everybody in the field seems only to talk about protecting training data. So why all the fuss there? Dont forget that the ultimate fact about ML is that the algorithm that does all of the learning is really just an instantiation of the data in machine runnable form!

So if your training set includes sensitive data, then by definition the machine you construct out of those data (using ML) includes sensitive information. And if your training set includes biased or regulated data, then by definition the machine you construct out of those data (using ML) elements includes biased or regulated information. And if your training set includes enterprise confidential data, then by definition the machine you construct out of those data (using ML) elements includes enterprise confidential information. And so on.

The algorithm is the data and becomes the data through training.

Apparently, the big focus the ML field puts on protecting training data has some merit. Not surprisingly, one of the main ideas for approaching the training data problem is to fix the training data so that it no longer directly includes sensitive, biased, regulated, or confidential data. At one extreme, you can simply delete those data elements out of your training set. Slightly less radical, but no less problematic is the idea of adjusting the training data in order to mask or obscure sensitive, biased, regulated, or confidential data.

Lets spend some time looking at that.

One of the hardest things to get straight in this new machine-learning paradigm is just who is taking on what risk. That makes the idea of where to place and enforce trust boundaries a bit tricky. As an example, we need to separate and understand not just operational data and training data as described above, but further determine who has (and who should have) access to training data at all.

And even worse, the question of whether any of the training data elements are biased, subject to protected class membership, protected under the law, regulated, or otherwise confidential, is an even thornier issue.

First things first. Somebody generated the possibly worrisome data in the first place, and they own those data components. So the data owner may end up with a bunch of data they are charged with protecting, such as race information or social security numbers or pictures of peoples' faces. That's the data owner.

More often than not, the data owner is not the same entity as the data scientist, who is supposed to use data to train a machine to do something interesting. That means that security people need to recognize a significant trust boundary between the data owner and the data scientist who trains up the ML system.

In many cases, the data scientist needs to be kept at arms length from the "radioactive" training data that the data owner controls. So how would that work?

Let's start with the worst approach to protecting sensitive training datadoing nothing at all. Or possibly even worse, intentionally doing nothing while you are pretending to do something. To illustrate this issue, we'll use Meta's claim about face-recognition data that was hoovered up by Facebook (now Meta) over the years. Facebook built a facial recognition system using lots of pictures of faces of its users. Lots of people think this is a massive privacy issue. (There are also very much real concerns about how racially biased facial-recognition systems are, but that is for another article.)

After facing privacy pressures over its facial recognition system, Facebook built a data transformation system that transforms raw face data (pictures) into a vector. This system is called Face2Vec, where each face has a unique Face2Vec representation. Facebook then said that it deleted all of the faces, even as it kept the huge Face2Vec dataset. Note that mathematically speaking, Facebook did nothing to protect user privacy. Rather, they kept a unique representation of the data.

One of the most common forms of doing something about privacy is differential privacy. Simply put, differential privacy aims to protect particular data points by statistically mungifying the data so that individually sensitive points are no longer in the data set, but the ML system still works. The trick is to maintain the power of the resulting ML system even though the training data have been borked through an aggregation and fuzzification process. If the data components are overly processed this way, the ML system cant do its job.

But if an ML system user can determine whether data from a particular individual was in the original training data (called membership inference), the data was not borked enough. Note that differential privacy works by editing the sensitive data set itself before training.

One system being investigated -- and commercialized -- involves adjusting the training process itself to mask sensitivities in a training dataset. The gist of the approach is to use the same kind of mathematical transformation at training time and at inference time to protect against sensitive data exposure (including membership inference).

Based on the mathematical idea of mutual information, this approach involves adding gaussian noise only to unconducive features so that a dataset is obfuscated but its inference power remains intact. The core of the idea is to build an internal representation that is cloaked at the sensitive feature layer.

One cool thing about targeted feature obfuscation is that it can help protect a data owner from data scientists by preserving the trust boundary that often exists between them.

Does all this mean that the problem of sensitive training data is solved? Not at all. The challenge of any new field remains: the people constructing and using ML systems need to build security in. In this case, that means recognizing and mitigating training data sensitivity risks when they are building their systems.

The time to do this is now. If we construct a slew of ML systems with enormous data exposure risks built right in, well, well get what we asked for: another security disaster.

View post:
Expert Insights: How to Protect Sensitive Machine-Learning Training Data Without Borking It - DARKReading

AI Can Offer Insight into Who Responds to Anti-Depressants – IEEE Spectrum

October 10, 2022 | Author: admin

I have been an otolaryngologist for more than two decades. My patients tell me they want more natural sound, more enjoyment of music, and most of all, better comprehension of speech, particularly in settings with background noisethe so-called cocktail party problem. For 15 years, my team at the University of Gttingen, in Germany, has been collaborating with colleagues at the University of Freiburg and beyond to reinvent the cochlear implant in a strikingly counterintuitive way: using light.

We recognize that todays cochlear implants run up against hard limits of engineering and human physiology. So were developing a new kind of cochlear implant that uses light emitters and genetically altered cells that respond to light. By using precise beams of light instead of electrical current to stimulate the cochlear nerve, we expect our optical cochlear implants to better replicate the full spectral nature of sounds and better mimic natural hearing. We aim to start clinical trials in 2026 and, if all goes well, we could get regulatory approval for our device at the beginning of the next decade. Then, people all over the world could begin to hear the light.

These 3D microscopic images of mouse ear anatomy show optical implants [dotted lines] twisting through the intricate structure of a normal cochlea, which contains hair cells; in deafness, these cells are lost or damaged. At left, the hair cells [light blue spiral] connect to the cochlear nerve cells [blue filaments and dots]. In the middle and right images, the bony housing of the mouse cochlea surrounds this delicate arrangement.Daniel Keppeler

Some 466 million people worldwide suffer from disabling hearing loss that requires intervention, according to the World Health Organization. Hearing loss mainly results from damage to the cochlea caused by disease, noise, or age and, so far, there is no cure. Hearing can be partially restored by hearing aids, which essentially provide an amplified version of the sound to the remaining sensory hair cells of the cochlea. Profoundly hearing-impaired people benefit more from cochlear implants, which, as mentioned above, skip over dysfunctional or lost hair cells and directly stimulate the cochlear, or auditory, nerve.

In the 2030s, people all over the world could begin to hear the light.

Todays cochlear implants are the most successful neuroprosthetic to date. The first device was approved by the U.S. Food and Drug Administration in the 1980s, and nearly 737,000 devices had been implanted globally by 2019. Yet they make limited use of the neurons available for sound encoding in the cochlea. To understand why, you first need to understand how natural hearing works.

In a functioning human ear, sound waves are channeled down the ear canal and set the ear drum in motion, which in turn vibrates tiny bones in the middle ear. Those bones transfer the vibrations to the inner ears cochlea, a snail-shaped structure about the size of a pea. Inside the fluid-filled cochlea, a membrane ripples in response to sound vibrations, and those ripples move bundles of sensory hair cells that project from the surface of that membrane. These movements trigger the hair cells to release neurotransmitters that cause an electrical signal in the neurons of the cochlear nerve. All these electrical signals encode the sound, and the signal travels up the nerve to the brain. Regardless of which sound frequency they encode, the cochlear neurons represent sound intensity by the rate and timing of their electrical signals: The firing rate can reach a few hundred hertz, and the timing can achieve submillisecond precision.

Hair cells in different parts of the cochlea respond to different frequencies of sound, with those at the base of the spiral-shaped cochlea detecting high-pitched sounds of up to about 20 kilohertz, and those at the top of the spiral detecting low-pitched sounds down to about 20 Hz. This frequency map of the cochlea is also available at the level of the neurons, which can be thought of as a spiraling array of receivers. Cochlear implants capitalize on this structure, stimulating neurons in the base of the cochlea to create the perception of a high pitch, and so on.

A commercial cochlear implant today has a microphone, processor, and transmitter that are worn on the head, as well as a receiver and electrodes that are implanted. It typically has between 12 and 24 electrodes that are inserted into the cochlea to directly stimulate the nerve at different points. But the saline fluid within the cochlea is conductive, so the current from each electrode spreads out and causes broad activation of neurons across the frequency map of the cochlea. Because the frequency selectivity of electrical stimulation is limited, the quality of artificial hearing is limited, too. The natural process of hearing, in which hair cells trigger precise points on the cochlear nerve, can be thought of as playing the piano with your fingers; cochlear implants are more equivalent to playing with your fists. Even worse, this large stimulation overlap limits the way we can stimulate the auditory nerve, as it forces us to activate only one electrode at a time.

The idea for a better way began back in 2005, when I started hearing about a new technique being pioneered in neuroscience called optogenetics. German researchers were among the first to discover light-sensitive proteins in algae that regulated the flow of ions across a cellular membrane. Then, other research groups began experimenting with taking the genes that coded for such proteins and using a harmless viral vector to insert them into neurons. The upshot was that shining a light on these genetically altered neurons could trigger them to open their voltage-gated ion channels and thus fire, or activate, allowing researchers to directly control living animals brains and behaviors. Since then, optogenetics has become a significant tool in neuroscience research, and clinicians are experimenting with medical applications including vision restoration and cardiac pacing.

Ive long been interested in how sound is encoded and how this coding goes wrong in hearing impairment. It occurred to me that stimulating the cochlear nerve with light instead of electricity could provide much more precise control, because light can be tightly focused even in the cochleas saline environment.

We are proposing a new type of implanted medical device that will be paired with a new type of gene therapy.

If we used optogenetics to make cochlear nerve cells light sensitive, we could then precisely hit these targets with beams of low-energy light to produce much finer auditory sensations than with the electrical implant. We could theoretically have more than five times as many targets spaced throughout the cochlea, perhaps as many as 64 or 128. Sound stimuli could be electronically split up into many more discrete frequency bands, giving users a much richer experience of sound. This general idea had been taken up earlier by Claus-Peter Richter from Northwestern University, who proposed directly stimulating the auditory nerve with high-energy infrared light, though that concept wasnt confirmed by other laboratories.

Our idea was exciting, but my collaborators and I saw a host of challenges. We were proposing a new type of implanted medical device that would be paired with a new type of gene therapy, both of which must meet the highest safety standards. Wed need to determine the best light source to use in the optogenetic system and how to transmit it to the proper spots in the cochlea. We had to find the right light-sensitive protein to use in the cochlear nerve cells, and we had to figure out how best to deliver the genes that code for those proteins to the right parts of the cochlea.

But weve made great progress over the years. In 2015, the European Research Council gave us a vote of confidence when it funded our OptoHear project, and in 2019, we spun off a company called OptoGenTech to work toward commercializing our device.

Our early proof-of-concept experiments in mice explored both the biology and technology at play in our mission. Finding the right light-sensitive protein, or channelrhodopsin, turned out to be a long process. Many early efforts in optogenetics used channelrhodopsin-2 (ChR2) that opens an ion channel in response to blue light. We used it in a proof-of-concept experiment in mice that demonstrated that optogenetic stimulation of the auditory pathway provided better frequency selectivity than electrical stimulation did.

In our continued search for the best channelrhodopsin for our purpose, we tried a ChR2 variant called calcium translocating channelrhodopsin (CatCh) from the Max Planck Institute of Biophysics lab of Ernst Bamberg, one of the world pioneers of optogenetics. We delivered CatCh to the cochlear neurons of Mongolian gerbils using a harmless virus as a vector. We next trained the gerbils to respond to an auditory stimulus, teaching them to avoid a certain area when they heard a tone. Then we deafened the gerbils by applying a drug that kills hair cells and inserted a tiny optical cochlear implant to stimulate the light-sensitized cochlear neurons. The deaf animals responded to this light stimulation just as they had to the auditory stimulus.

The optical cochlear implant will enable people to pick out voices in a busy meeting and appreciate the subtleties of their favorite songs.

However, the use of CatCh has two problems: First, it requires blue light, which is associated with phototoxicity. When light, particularly high-energy blue light, shines directly on cells that are typically in the dark of the bodys interior, these cells can be damaged and eventually die off. The other problem with CatCh is that its slow to reset. At body temperature, once CatCh is activated by light, it takes about a dozen milliseconds to close the channel and be ready for the next activation. Such slow kinetics do not support the precise timing of neuron activation necessary to encode sound, which can require more than a hundred spikes per second. Many people said the kinetics of channelrhodopsins made our quest impossiblethat even if we gained spectral resolution, wed lose temporal resolution. But we took those doubts as a strong motivation to look for faster channelrhodopsins, and ones that respond to red light.

We were excited when a leader in optogenetics, Edward Boyden at MIT, discovered a faster-acting channelrhodopsin that his team called Chronos. Although it still required blue light for activation, Chronos was the fastest channelrhodopsin to date, taking about 3.6 milliseconds to close at room temperature. Even better, we found that it closed within about 1 ms at the warmer temperature of the body. However, it took some extra tricks to get Chronos working in the cochlea: We had to use powerful viral vectors and certain genetic sequences to improve the delivery of Chronos protein to the cell membrane of the cochlear neurons. With those tricks, both single neurons and the neural population responded robustly and with good temporal precision to optical stimulation at higher rates of up to about 250 Hz. So Chronos enabled us to elicit near-natural rates of neural firing, suggesting that we could have both frequency and time resolution. But we still needed to find an ultrafast channelrhodopsin that operated with longer wavelength light.

We teamed up with Bamberg to take on the challenge. The collaboration targeted Chrimson, a channelrhodopsin first described by Boyden thats best stimulated by orange light. The first results of our engineering experiments with Chrimson were fast Chrimson (f-Chrimson) and very fast Chrimson (vf-Chrimson). We were pleased to discover that f-Chrimson enables cochlear neurons to respond to red light reliably up to stimulation rates of approximately 200 Hz. Vf-Chrimson is even faster but is less well expressed in the cells than f-Chrimson is; so far, vf-Chrimson has not shown a measurable advantage over f-Chrimson when it comes to high-frequency stimulation of cochlear neurons.

This flexible micro-LED array, fabricated at the University of Freiburg, is wrapped around a glass rod thats 1 millimeter in diameter. The array is shown with its 144 diodes turned off [left] and operating at 1 milliamp [right]. University of Freiburg/Frontiers

Weve also been exploring our options for the implanted light source that will trigger the optogenetic cells. The implant must be small enough to fit into the limited space of the cochlea, stiff enough for surgical insertion, yet flexible enough to gently follow the cochleas curvature. Its housing must be biocompatible, transparent, and robust enough to last for decades. My collaborators Ulrich Schwarz and Patrick Ruther, then at the University of Freiburg, started things off by developing the first micro-light-emitting diodes (micro-LEDs) for optical cochlear implants.

We found micro-LEDs useful because theyre a very mature commercial technology with good power efficiency. We conducted severalexperiments with microfabricated thin-film micro-LEDs and demonstrated that we could optogenetically stimulate the cochlear nerve in our targeted frequency ranges. But micro-LEDs have drawbacks. For one thing, its difficult to establish a flexible, transparent, and durable hermetic seal around the implanted micro-LEDs. Also, micro-LEDs with the highest efficiency emit blue light, which brings us back to the phototoxicity problem. That's why were also looking at another way forward.

Instead of getting the semiconductor emitter itself into the cochlea, the alternative approach puts the light source, such as a laser diode, farther away in a hermetically sealed titanium housing. Optical fibers then bring the light into the cochlea and to the light-sensitive neurons. The optical fibers must be biocompatible, durable, and flexible enough to wind through the cochlea, which may be challenging with typical glass fibers. Theres interesting ongoing research in flexible polymer fibers, which might have better mechanical characteristics, but so far, they havent matched glass in efficiency of light propagation. The fiber-optic approach could have efficiency drawbacks, because wed lose some light when it goes from the laser diode to the fiber, when it travels down the fiber, and when it goes from the fiber to the cochlea. But the approach seems promising, as it ensures that the optoelectronic components could be safely sealed up and would likely make for an easy insertion of the flexible waveguide array.

Another design possibility for optical cochlear implants is to use laser diodes as a light source and pair them with optical fibers made of a flexible polymer. The laser diode could be safely encapsulated outside the cochlea, which would reduce concerns about heat, while polymer waveguide arrays [left and right images] would curl into the cochlea to deliver the light to the cells.OptoGenTech

As we consider assembling these components into a commercial medical device, we first look for parts of existing cochlear implants that we can adopt. The audio processors that work with todays cochlear implants can be adapted to our purpose; well just need to split up the signal into more channels with smaller frequency ranges. The external transmitter and implanted receiver also could be similar to existing technologies, which will make our regulatory pathway that much easier. But the truly novel parts of our systemthe optical stimulator and the gene therapy to deliver the channelrhodopsins to the cochleawill require a good amount of scrutiny.

Cochlear implant surgery is quite mature and typically takes only a couple of hours at most. To keep things simple, we want to keep our procedure as close as possible to existing surgeries. But the key part of the surgery will be quite different: Instead of inserting electrodes into the cochlea, surgeons will first administer viral vectors to deliver the genes for the channelrhodopsin to the cochlear nerve cells, and then implant the light emitter into the cochlea.

Since optogenetic therapies are just beginning to be tested in clinical trials, theres still some uncertainty about how best to make the technique work in humans. Were still thinking about how to get the viral vector to deliver the necessary genes to the correct neurons in the cochlea. The viral vector weve used in experiments thus far, an adeno-associated virus, is a harmless virus that has already been approved for use in several gene therapies, and were using some genetic tricks and local administration to target cochlear neurons specifically. Weve already begun gathering data about the stability of the optogenetically altered cells and whether theyll need repeated injections of the channelrhodopsin genes to stay responsive to light.

Our roadmap to clinical trials is very ambitious. Were working now to finalize and freeze the design of the device, and we have ongoing preclinical studies in animals to check for phototoxicity and prove the efficacy of the basic idea. We aim to begin our first-in-human study in 2026, in which well find the safest dose for the gene therapy. We hope to launch a large phase 3 clinical trial in 2028 to collect data that well use in submitting the device for regulatory approval, which we could win in the early 2030s.

We foresee a future in which beams of light can bring rich soundscapes to people with profound hearing loss or deafness. We hope that the optical cochlear implant will enable them to pick out voices in a busy meeting, appreciate the subtleties of their favorite songs, and take in the full spectrum of soundfrom trilling birdsongs to booming bass notes. We think this technology has the potential to illuminate their auditory worlds.

From Your Site Articles

Clustering of trauma patients based on longitudinal data and the application of machine learning to predict recovery | Scientific Reports – Nature.com

October 10, 2022 | Author: admin

Principal component analysis

The development of a supervised machine learning model that can predict the recovery profile of trauma patients requires labelled data. Since the BIOS data set does not contain patient classifications based on recovery, the first step is to cluster patients based on similarity across the different outcome variables that represent the health condition (unsupervised learning). The topic of similarity-based clustering has been investigated intensively, both from a statistical modeling point of view as well as using Machine/Deep Learning approaches48. Preliminary analyses showed that although there is some correlation between the variables measuring recovery, based on clinical expertise it might make sense to separate the recovery variables dealing with physical status from those dealing with psychological function. In the frame of this study, we thus focus on four different cases for the extraction of the clusters: Physical health (with and without pre-trauma scores), psychological health, and general health.

For the case of Physical Health, we implement two cases (i) longitudinal profiles post-injury with four variables namely EQ-5D, EQ-VAS, HUI2 and HUI3 and (ii) longitudinal profiles including pre-injury data with two variables EQ-5D and EQ-VAS (for which pre-injury values Pre-injury EQ-5D and Pre-injury EQ-VAS were available). For Psychological Health the variables HDSA, HDSD and IES are used. Finally, for the case of General Health, a combination of Physical (no pre-injury values) and Psychological variables is applied (EQ-5D, EQ-VAS, HUI2 and HUI3, HDSA, HDSD and IES). For General Health, in total forty-two components are present since we have seven variables for six time frames. To investigate how Physical Health variables correlate with Psychological variables PCA was carried out to visualize the correlations between variables and inspect their loading on the principal components.

The first three components explain more than 60% of the total variance of the forty-two components (see Fig. S4 in the Supplementary information). Additionally, for the first three components, the biggest increase in the cumulative explained variance is observed. For this reason, we extracted the first three components for further analysis. Figure1 displays the PCA biplot for the first and third component. Different color codes are used to represent the ten patient clusters derived from the set of General Health variables by the kml3d method. The plot shows that the first dimension represents the general health condition of the patients. Positive values on dimension one represent good health of patients two years after trauma while negative values point to poor health. The centers of the clusters move consistently from positive to negative values as we move from the first clusters A, B and C (high initial health and high recovery) to the last clusters J, I and H (low initial health and low recovery) (Fig.1).

PCA biplot for the set of General Health variables with indication of the ten clusters obtained with kml3d.

Moreover, we observe that Psychological and Physical Health vectors correlate with each other since they generally point to the same direction for the first dimension. The third dimension represents time as positive values point to variables for the first and the second time frame. As we move to negative values of component three we observe the last three time frames. The correlation of the general health variables is stronger (vectors overlap with each other) as we move from the first to the last time frames. (The second dimension splits the physical from the psychological variables, see Fig. S4 in the Supplementary information).

For the clusters obtained with kml3d, the optimum number is calculated based on the gap statistic using the "Nbclust" library in R45. The gap statistic provides us with the optimal number of clusters per set of variables (Fig.2). For Physical Health, the optimal number of clusters is eight, for Psychological Health it is nine, for General Health it is ten and finally for Physical Health with pre-injury values it is eight. For the clusters obtained with HDclassif and Deepgmm, a grid search was executed per set of variables in combination with the BIC to determine the optimal number of clusters and setting of the parameters; the results of this are presented in Table 1.

Optimum number of clusters with kml3d for the four different cases of variables and k-means.

In general, the number of optimum clusters reduced when we apply HDclassif and Deepgmm compared with kml3d. Additionally, clusters obtained with kml3d are generally more balanced (majority baseline is at maximum of 26.15%). On the other hand, unbalanced clusters (except for the case of Physical Health) are obtained when we apply the Deepgmm clustering method (high majority baselines).

For predicting the outcome class of the patients, we use the labels generated in the clustering step as the target for prediction in a number of supervised machine learning models. In this following example, we focus the model comparison step for the prediction of the six class labels derived from clustering the set of Physical Health variables including pre-injury values with the HDclassif method. We used Logistic Regression, Random Forest and XGBoost as models with different settings for under- or oversampling and hyperparameters. All models were compared under 5-fold cross validation, and we report the mean f1 macro and the 95% CI for accuracy for this example model comparison step in Table 2. We report next to accuracy the f1 macro score since we deal with imbalanced data sets where all the classes are equally important. It is clear that over-sampling has a positive impact on the classification task resulting in higher accuracy and that the Random Forest and XGBoost algorithms outperform logistic regression in this case.

For this reason, Random Forest with over-sampling is the algorithm that we used for the prediction of the classes derived from all clustering attempts using the three clustering methods (Table 3). The best classification results are observed for the clusters obtained with the Deepgmm method. However, the majority baselines for these cluster solutions are high (from 61.02 to 84.70%) meaning that clusters are highly unbalanced. A more detailed methodology for the evaluation of the clusters (clinical sensibleness) based on medical expertise is described in the next section.

In order to get a thorough understanding about the prediction, a technique called Boruta is applied to the prediction models47. Boruta is a feature selection algorithm, implemented as a wrapper algorithm around Random Forest. In Table 4, the prediction accuracy is presented both with all (26) predictors and only with the important predictors extracted with Boruta for the case of General Health. For kml3d and HDclassif, the same seven predictors are highlighted as important. For the case of Deepgmm, the same predictors are noted as important predictors excluding BMI and including predictors such as Category accident, Education level, Traumatic brain injury, Gender and Pre-injury cognition. As can be seen, applying Boruta feature selection did not impair accuracies, leading to simpler models that did not compromise on classification accuracy.

In the previous section, models with high accuracy were developed for the classification of patients. Specifically, clusters derived from Deepgmm are predicted with high accuracy applying Random Forest and over-sampling. Since the obtained clusters cannot be directly evaluated in terms of representing observable ground-truth classes, the strategy to arrive at sensible and functional models is to combine several quality indicators based on statistical criteria, machine learning metrics, and clusters quality assessment based on medical expertise (clinical sensibleness) in relation to known risk factors for recovery. An example of the applied clusters quality assessment is presented in this section for the clusters obtained with three different methods. For illustration purposes we selected three cases which represent highly, medium and poor sensible clustering (Table 5).

For the case of General Health using the HDclassif method the optimal number of clusters is six. In Table 5 the descriptive statistics per cluster are presented. The order of the clusters is defined from the younger to the older patients. As can be seen, there is a trend for the age of the patients to increase across clusters in this highly sensible model (+++). Specifically, for patients who belong to the first cluster (cluster 1) the mean age is around fifty-eight while for patients who belong to the last cluster (cluster 6) the mean age is around seventy-five. Looking at frailty and comorbidities we observe that older patients are characterized by more comorbidities and higher frailty. Moreover, young patients with less frailty are admitted in the hospital for fewer days and their severity score is also lower compared with patients who are older with more frailty. Additionally, exploring the gender distribution of the clusters we observe that the percentage of females increases as we move from the first to the last clusters. Looking at hip fracture injuries, the clusters quality assessment reveals that the last clusters contain a higher percentage of patients who suffer from this known risk factor for poor recovery. The medium and low clinically sensible models do not recapitulate these demographic risk-factor differences as clearly across clusters.

Recovery of the patients is measured based on various parameters. For two parameters, namely EQ-5D and EQ-VAS, we also have pre-injury estimated baseline values. These variables describe the self-reported physical condition of the patients before their injury. These values can thus be used as a baseline for the analysis of patient recovery (Top two graphs in Fig.3). As can be seen from the two graphs, EQ-5D and EQ-VAS show a dip from baseline (set at 100%) and show recovery over time. Patients who belong to the first clusters (1, 2) recover almost completely while for patients of the last clusters (5, 6) recovery is about 6084% depending on the variable.

The two graphs at the top present recovery based on EQ-VAS and EQ-5D for the case of General Health with HDclassif. The two graphs at the bottom depict psychological condition (high values indicate high stress and anxiety) of various clusters after the injury.

Psychological condition is also relevant for the recovery of the patients and is plotted in the bottom graphs of Fig.3. Psychological condition is measured with three parameters namely HDSA (Anxiety), HDSD (Depression) and IES. The same trend over time after the accident can be observed for these parameters. More particular, patients who belong to the first clusters (high recovery) appear to have low levels of depression and anxiety. For the first three clusters (1, 2 and 3) the level of stress decreases over time. On the other hand, for clusters 4, 5, and 6 the level of stress and anxiety remains high for a month and then start decreasing.

According to medical experience, the clusters obtained for General Health Case using the HDclassif method meet the expectations and agree with the prototypical cases observed at the hospital. Especially, the group of old females with high frailty and with a hip fracture is a characteristic group observed at the hospital and typically has low recovery. On the other hand, younger male patients with less comorbidities, low severity score and less days admitted to the hospital recover completely and appear to have low levels of stress and anxiety.

For the selection of a rational and functional model that makes clinical sense, we thus implemented a cluster quality assessment as described in the previous paragraphs for each cluster model case. As a reference we use the case of General Health with HDclassif method (highly sensible). The results are presented in Table 3. Based on the clusters quality assessment each case is categorized on clinical sensibleness either as Poorly sensible (+) or as Medium sensible (++) or as Highly sensible (+++). Highly sensible clusters are those cases where the clusters quality assessment reveals discrete clusters with the same trends and characteristics as the reference case (General Health with HDclassif method) matching clinical experience. On the contrary, when we have clusters that are not discrete or without the characteristics of the reference group then the model is categorized as less adequate. This is the case for example for the clusters obtained for Psychological Health with Deepgmm (Table 5). Descriptive statistics of the clusters obtained for this case reveal that clusters are not discrete and do not follow the characteristic trends for frailty, comorbidities, severity score or days admitted in the hospital. Gender and hip fracture do not follow the trend of the reference case.

Performing clusters quality assessment together with medical experts, we discovered that there are cases where clusters partly match with the clusters of the reference case. In this case not all the clusters are discrete. There are clusters which appear similar properties. However, some of the trends of the clusters match with the trends of the reference group. In Table 5 an example of medium sensible case is presented for the case of Physical Health (pre-injury) and the clustering technique of kml3d. For this case although we observe trends between the clusters for the different variables, there are clusters such as (B, C) and (A, E) who are not discrete and do not follow the general trend of the reference clusters More particular, even though cluster C contains patients with slightly lower Age than cluster B, the mean values of Frailty and Comorbidities are higher.

A supplementary method to quantify the separability of the obtained clusters is to execute a MANOVA. More particular, non-parametric MANOVA (using the function adonis from library "vegan"49) is executed for the clusters of all cases on the variables of Age, Frailty, Comorbidities, Injury severity score, Pre-injury EQ-5D, and T6-EQ-5D. We decided to execute a non-parametric MANOVA since the assumptions for running MANOVA (homogeneity of the variances and normality within the groups) were not met for our data. Assumptions are examined using the function assumptions manova. The non-parametric MANOVA revealed that there was a strong relation between the value of the F statistic and the sensibility of the clusters. More precisely, F values between 79.71 and 101.45 (separable clusters, very low p-values) are obtained for the highly sensible models. For medium sensible clusters F is between 5.99 and 8.32 while for inadequate clusters F is between 1.21 and 2.98. For the non-sensible clusters, the difference between clusters is not significant, showing p-values larger than the chosen threshold of 5%. It is remarkable that for the case of Physical Health with Deepgmm method, non-parametric MANOVA reveals that there is a statistically significant difference between the obtained clusters, F(35, 3880)=101.45, p<103. On the contrary, for the case of Psychological Health with Deepgmm, non-parametric MANOVA indicates that the separability of the clusters is not statistically significant F(35, 3880)=1.21, p=0.30.

Further evaluation of the models is performed by using a graphical method: plotting the t-distributed stochastic neighbor embedding (t-SNE) graphs. In the Supplementary information the t-SNE graphs of two extreme cases, namely General Health with kml3d and 10 clusters with high clinical sensibleness and Psychological Health with Deepgmm with 6 clusters with low clinical sensibleness, are presented (see Fig. S5 and Fig. S6 in the Supplementary information). In the case of General Health with kml3d, t-SNE visualisation shows discrete clusters in the two-dimensional space. On the contrary, for Psychological Health with Deepgmm, high interference between the groups is observed.

From Table 3, we observe that for General Health, the best model is achieved with the HDclassif method. The accuracy of this model is almost 74% while the clusters quality assessment indicates that the obtained clusters are sensible. For the case of Physical Health, the best model with high accuracy (91.30%) and sensible clusters is derived using the Deepgmm method. Cluster quality assessment of clusters obtained with the HDclassif method for Physical Health with pre-injury measurements reveals that clusters are highly sensible, however, accuracy is much lower (at 69.12%) compared with Deepgmm. Another observation has to do with the case of Psychological Health. Applying variables which are related only to the psychological condition of the patients do not lead to sensible (+++) clusters for any method, suggesting that these outcome measures are not related to traditional risk factors for physical recovery, but capture a different dimension.

Original post:
Clustering of trauma patients based on longitudinal data and the application of machine learning to predict recovery | Scientific Reports - Nature.com

Fast and scalable search of whole-slide images via self-supervised deep learning – Nature.com

October 10, 2022 | Author: admin

SISH

SISH is a histology-image search pipeline that addresses the scalability issues of speed, storage and pixel-wise label scarcity. While image-level class labels or annotation for whether a pair or triplet of images are similar/dissimilar has often been leveraged to explicitly learn an embedding space that captures semantic similarity between images, or to identify key points before retrieval50,51, it is difficult to directly apply such techniques to WSI image search due to the high dimensionality of large WSIs and the lack of patch-level annotation. Instead, SISH builds upon a set of preprocessed mosaics from WSIs without pixel-wise or ROI-level labels to reduce storage and labelling costs, by relying on indices learned via self-supervised learning and pretrained embeddings. SISH scales with a constant search speed, regardless of the size of the database, by taking advantage of the benefits of discrete latent codes from a VQ-VAE, and using guided search and ranking algorithms. We present these essential components of SISH in this section and provide a pictorial illustration for the methods described in Fig. 2. For clarity, we have summarized all symbols used in the following text in Supplementary Table 18.

VQ-VAE41 is a variant of a Variational AutoEncoder (VAE) that introduces a training objective that allows for discrete latent codes. Let ({{{boldsymbol{e}}}}in {{mathbb{R}}}^{Ktimes D}) be the latent space (that is, codebook) where K is the number of discrete codewords and D is the dimension of the codewords. We set K=128 and D=256 in our experiments. To decide the codeword of the given input, an encoder q encodes input x as ze(x). The final codeword zq(x) of x and the training objective function are given by

$${{{{boldsymbol{z}}}}}_{q}(x)={{{{boldsymbol{e}}}}}_{k},{{{rm{where}}}} k={{{{rm{argmin}}}}}_{j}leftVert {{{boldsymbol{{z}}}_{e}}}({{{boldsymbol{x}}}})-{{{{boldsymbol{e}}}}}_{j}rightVert ,$$

(1)

$$log p({{{boldsymbol{x}}}}| {{{{boldsymbol{z}}}}}_{q}({{{boldsymbol{x}}}}))+leftVert {{{rm{sg}}}}[{{{boldsymbol{{z}}}_{e}}}({{{boldsymbol{x}}}})]-{{{boldsymbol{e}}}}rightVert +alpha leftVert {{{boldsymbol{{z}}}_{e}}}({{{boldsymbol{x}}}})-{{{rm{sg}}}}[{{{boldsymbol{e}}}}]rightVert ,$$

(2)

where is a hyperparameter and sg denotes the stop gradient operation. A stop gradient operation acts as the identity function during the forward pass while having zero gradient during the backward pass. The first term in the objective function optimizes the reconstruction of the encoder and decoder, the second term is used to update the codebook, and the third term is used to prevent the encoders output from diverging too far from the latent space. The architecture of our VQ-VAE model is shown in detail in Supplementary Fig. 4. We re-ordered the codebook on the basis of the value of the first principal component and changed the latent code accordingly as we found that the re-ordered codebook can provide a more semantic representation of the original input image (see Supplementary Fig. 5).

We show how each patch i in the mosaic of a WSI can be represented by a tuple (pi, hi) composed of a patch index pi and a patch texture feature hi. To get pi, we encode and re-map the latent code zi from the encoder and re-ordered codebook from the VQ-VAE. The index pi is determined by the following equations:

$${{{{boldsymbol{z}}}}}_{i,1}={{{rm{AVGPOOL}}}}(2,2)({{{{boldsymbol{z}}}}}_{i})$$

(3)

$${{{{boldsymbol{z}}}}}_{i,2}={{{rm{AVGPOOL}}}}(2,2)({{{{boldsymbol{z}}}}}_{i,1})$$

(4)

$${{{{boldsymbol{z}}}}}_{i,3}={{{rm{AVGPOOL}}}}(2,2)({{{{boldsymbol{z}}}}}_{i,2})$$

(5)

$${p}_{i}={{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{i,1})+{{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{i,2})times 1{0}^{6}+{{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{i,3})times 1{0}^{11}$$

(6)

$${{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{{{{boldsymbol{i,1}}}}})in [0,130048],{{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{{{{boldsymbol{i,2}}}}})in [0,32512],{{{rm{SUM}}}}({{{{boldsymbol{z}}}}}_{{{{boldsymbol{i,3}}}}})in [0,8128]$$

(7)

To convert the information in the latent code from higher to lower resolution, we apply a series of 22 average pooling. We then take the summation to aggregate the latent code in each resolution, as the summation operator has better expressive power than the mean or maximum52. We get the final integer index by taking the summation of the information aggregated in each resolution and multiplying it by 100, 106 and 1011, respectively. The intuition behind choosing the power is to keep the information of the latent code in each resolution (that is, zi,1, zi,2 and zi,3) separate. For example, multiplying Sum(zi,2) by 106 separates the feature in the second layer from Sum(zi,1) since the maximum of the latter is 130,048. Likewise, multiplying Sum(zi,3) by 1011 separates the feature in the last layer from the previous two. We insert each pi into the vEB tree for fast search. We apply this process to WSIs in the datasets to build our databases. Note that the time complexity of all operations in the vEB tree is (O(log log (M))). On the basis of the properties of the vEB tree, M can be determined by

$$M={2}^{x} > max ({p}_{i}),,$$

(8)

where x is the minimum integer that satisfies the inequality. Since our codebook size ranges from 0 to 127, we can determine the maximum summation Sum(z) in each level. Solving the inequality, we find that the minimum M that satisfies the inequality is M=1,125,899,906,842,624. Because M is a constant that only depends on the index generation pipeline, our search performance is O(1). One limitation of using vEB is that it has a large space complexity O(M) where M depends on the size of the codebook and the dynamic range of the index used for search. M remains fixed and does not scale with the number of data points (WSIs or patches) in the database. To get hi, we use DenseNet121 to extract features from the 1,0241,024 patch at 20, then follow the algorithm proposed in ref. 36 to binarize it (that is, starting from ; if the next value is smaller than the current one, the current value 0 is assigned, and 1 is assigned otherwise).

In addition to creating the tuple to represent the patch, we also make a hash table H with pi as key and the metadata of the patch as value. The metadata include the texture feature hi, the name of the slide associated with the patch, the coordinates on the slide from which the patch is cropped, the file format of the slide and the diagnosis of the slide. Note that different patches could share the same key. In this case, the value is a list that stores the metadata for each patch. If the size of the search database is significantly large, which is expected to be the case for most practical real-world databases, the search speed would be greater than pre- and post-processing steps. When running a fixed number of iterations, the K-means clustering algorithm (Lloyds algorithm) has time complexity O(BKIC) where B is the number of patches in a WSI, K is the number of cluster centroids, I is the number of iterations and C is the dimension of each input data point. For fixed I, K and C, the initial clustering steps of mosaic construction is O(B). To obtain the final mosaic, a fixed percentage (e.g. 5%) of patches are sampled from each cluster, and hence the resulting mosaic varies from slide to slide with size B = 0.05 B. During retrieval, the number of total candidates proposed (before ranking) is T (ksucc + kpred) B (see the next section for the definition of T, ksucc and kpred). For ranking, the complexity is O(B). Therefore, given fixed ksucc, kpred and T, the time complexity of retrieval is O(B). Note that since the size of a WSI is capped by the physical size of the glass slide and the tissue specimen, for a fixed patch size, we can always pick a reasonable constant Bmax to upper bound the maximum B in the database and in incoming query slides. Therefore, the entire workflow has a theoretical constant time complexity of O(1). In real-world scenarios where we expect the size of the database to scale to hundreds of thousands or millions of slides, the time complexity of retrieval will dominate over other steps such as mosaic generation and ranking if we do not use an O(1) search algorithm and it instead scales with O(n) or O(nlogn), where n is the size of the database. However, we note that while in most practical scenarios with increasingly large databases, the size of the WSI database (n) would be larger than the size of the number of patches in the query slide (B); in rare cases where the size of the database is very small, such that average B is not negligible compared to n, while the search operation will continue to have a constant O(1) complexity, the speed of the overall pipeline may be limited by the mosaic generation O(Bmax). Mosaic generation can also be completed before case review, further improving search speeds.

For clarity, we use mi to denote the patch index in the mosaic of the query slide to distinguish those in the database. Given a query slide I represented as I={(m1,h1),(m2,h2),,(mk,hk)} with k patches, where each tuple is composed of the index of the patch mi and its texture features hi, we apply guided-search to each tuple and return the corresponding results ri. The output takes the form of RI={r1,r2,,rk}. Each ri={(pi1,i1),(pi2,i2),,(pin,in)}, a set of tuples consisting of the indices of similar patches (pi1,pi2,,pin) and their associated metadata (i1,i2,,in). ij includes all metadata associated with the j-th patch plus the Hamming distance between hi and hj. A visual illustration is shown in Fig. 2.

The drawback to using only mi for the query is that the current patch index is sensitive to minor changes in zi,3. For example, a patch that differs from another by 1 incurs a 1011 difference in index, putting the two patches far from each other in the vEB tree. To address this issue, we create a set of candidate indices mi,c+ and mi,c along with the original mi by adding and subtracting an integer C for T times from Sum(zi,3). We call helper functions forward-search and backward-search to search the neighbour indices in mi,c+ and mi,c, respectively. Both functions include only those neighbouring indices whose Hamming distance from the query hi is smaller than a threshold, h. The details of these algorithms are presented in Algorithms 1 through 3.

Algorithm 1 Guided Search Algorithm

Hhash table Hash table with patch index as key and metadata as value

C,T501011,10 Integer and number of times for addition and subtraction

h128 Threshold of the Hamming distance between query patch index and the neighbor

ksucc,kpred375 Number of time to call vEB.Successor() and vEB.Predecessor()

Function GUIDED-SEARCH(mi,hi,C,T,h,kpred,ksucc,H,vEB)

mi,c+,mi,c,results{},{},{}

V{}

mi,c+.insert(mi)

for t1,2,...,T do

mtmp+,mtmp-mi+tC,mitC

mi,c+.insert(mtmp+)

mi,c.insert(mtmp-)

results+,VFORWARD-SEARCH(mi,c+,ksucc,h,V,H,vEB)

resultsBACKWARD-SEARCH(mi,c,kpred,h,V,H,vEB)

results.insert(results+)

results.insert(results)

resultsSORT-ASCENDING(results,key=results.hammingdistance)

return results

Our ranking function Ranking (Algorithm 4) takes the results RI={r1,r2,,rk} from Guided-Search as input. The output is the top-k similar slides given the query slide I. We set k equal to 5 for all experiments, except for anatomic site retrieval where k is equal to 10. The intuition behind Ranking is to find the most promising patches in RI on the basis of the uncertainty. It relies on three helper functionsWeighted-Uncertainty-Cal (Algorithm 5), Clean (Algorithm 6) and Filtered-By-Prediction (Algorithm 7).

Weighted-Uncertainty-Cal (Algorithm 5) takes RI as input and calculates the uncertainty for each ri by computing their entropy (that is, frequency of occurrence of slide labels). The lower the entropy, the less uncertain the patch and vice versa. The output is the entropy of ri, along with records that summarize the diagnosis occurrences and Hamming distance of each element in ri. The disadvantage of counting the occurrences naively in the entropy calculation is that the most frequent diagnosis in the anatomic site dominates the result and therefore downplays the importance of others. We introduce a weighted occurrence approach to address this issue. The approach counts the diagnosis occurrences by considering the percentage of the diagnosis in the given site and the diagnosiss position in the retrieval results. It calculates the weight of each diagnosis in the anatomic sites by the reciprocal of the number of diagnosis. We normalize the weights such that their summation is equal to a constant N. A diagnosiss final value in ri is the normalized weight of the diagnosis multiplied by the inverse of position where the diagnosis appears in ri. Therefore, the same diagnosis can have different weighted occurrences because of its position in ri. As such, less frequent diagnoses and those with lower Hamming distance (that is, close to the beginning of the retrieval results) gain more importance in the ranking process. As illustrated in Fig. 2, we summarize RI with three metadata values, Slb, Sm and Sl, to facilitate the subsequent processes. Specifically, Sm is a list that stores tuples of form (index of ri, entropy, Hamming distance info in ij, length of ri), Sl is an array that only stores the length of ri and Slb is a nested dictionary that stores the disease occurrences in ri.

Algorithm 2 Forward Search Algorithm

functionForward-Search(mi,c+,ksucc,h,V,H,vEB)

res+{}

for i+ in mi,c+ do

succ_cnt,succprev0,i+

whilesucc_cnt

succvEB.Successor(succprev)

ifsuccV or succ is empty then

break

else ifH[succ].len()==0 then

// The case when the patient is identical to query slide I

succprevsucc

continue

else

// Find the patch with smallest Hamming distance in the same key

distj,jArgminj(Hamming-Distance(hi,H[succ]))

ifdistj

V.insert(succ)

metaH[succ][j]

res+.insert((distj,meta))

succ_cnt,succprevsucc_cnt+1,succ

returnres+,V

Algorithm 3 Backward Search Algorithm

functionBackward-Search(mi,c,kpred,h,V,H,vEB)

res{}

fori in mi,cdo

pred_cnt,predprev0,i

whilepred_cnt

predvEB.Predecessor(predprev)

ifpredV or pred is emptythen

break

else ifH[pred].len()==0then

// The case when the patient is identical to query slide I

predprevpred

continue

else

// Find the mosaic with smallest Hamming distance in the same key

distj,jArgminj(Hamming-Distance(hi,H[pred]))

ifdistj

V.insert(pred)

metaH[pred][j]

res.insert((distj,meta))

pred_cnt,predprevpred_cnt+1,pred

returnres

Algorithm 4 Results Ranking Algorithm

function RANKING(Rs, D_inv, N, K)

if Rs is empty then return

D_invNORMALIZE(D_inv,N) Normalize the reciprocal of diagnosis count so that the sum is equal to N. N=10 for the fixed site and N=30 for the anatomic site experiments, respectively.

Slb,Sm{},{}

Sl{}

for each patchs results ri in RS do

if ri is not empty then

Ent,label_cnt,distWEIGHTED-UNCERTAINTY-CAL(ri,D_inv)

CalypsoAI brings testing and validation to ML models used by the TSA – VentureBeat

October 10, 2022 | Author: admin

Learn how your company can create applications to automate tasks and generate further efficiencies through low-code/no-code tools on November 9 at the virtual Low-Code/No-Code Summit. Register here.

Artificial intelligence (AI) models are increasingly finding their way into critical aspects of enterprise use cases and broader adoption throughout the world.

One area where AI is finding a home is in the Transport Security Administration (TSA), one of whose responsibilities is screening baggage at airports across the U.S. An initiative currently underway within the Department of Homeland Securitys (DHS) Science and Technology Directorate, a program known as Screening at Speed (SaS), will, among other efforts, implement AI to help accelerate the baggage screening process.

Part of developing this screening system is testing and validating the AI models integrity, in terms of both reliability and the ability to withstand potential adversarial AI attacks. Thats where DHS is making use of CalypsoAIs VESPR Validate technology.

Were really focused on testing and validation of machine learning (ML) models so that they can get safely and securely deployed, CalypsoAI CEO Neil Serebryany told VentureBeat.

Low-Code/No-Code Summit

Join todays leading executives at the Low-Code/No-Code Summit virtually on November 9. Register for your free pass today.

A Gartner research survey released in August found that only half of all AI models built actually make it into production.

There are multiple reasons for this, including issues with testing and validation. According to Serebryany, testing and validation for AI models must consider both human and technical factors. In order to help someone get the confidence they need to deploy a model into production, there is a need to solve for the human side. Human factors include the ability to communicate information about where the model works, where it doesnt work, and what its vulnerabilities are. On the technical side, Serebryany said that there is a need to help make the models as resilient and robust as possible.

Before starting CalypsoAI, Serebryany had worked in the government, where he noticed a growing focus on machine learning (ML). What he saw time and again were security challenges, including the need to make sure that adversarial machine learning attacks dont negatively impact a model.Adversarial ML attacks use various techniques to deceive models into generating the desired outcome.

The need for AI testing and validation as well as protection against adversarial AI extends beyond government use cases.

Serebryany said that his firm has seen growing enterprise demand recently. In fact, in July, Gartner named the company a Cool Vendor for its scaling AI in the Enterprise capabilities.

Organizations are trying to systematize how they understand the risk of their machine learning models and have a way of actually testing that risk in order to be able to validate their models, Serebryany said.

He expects that the need to test and validate AI will become part of many organizations audit practices to help ensure regulatory compliance. For example, he noted that the EU is starting to introduce regulations for AI compliance that enterprises will need to deal with.

Serebryany says that his company is also seeing insurance companies who want to start insurance AI models. Insurance companies need to be able to understand the performance of those models against a real-world test set of conditions.

Serebryany explained that his companys technology can fit into different parts of an AI development workflow.

CalypsoAI has a software development kit (SDK) that can work with an organizations Jupyter notebook-driven machine learning processes. Alternatively, CalypsoAI can be just an independent testing and validation step along the way.

Serebryany explained that CalypsoAI can test a model without all the training data. Using a subset of data, CalypsoAI runs the model through a series of adversarial attacks and real-world scenarios.

A lot of the testing challenges are around identifying the exact conditions folks want to deployinto and helping them understand what model will actually be performing in that condition set, he said.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.

See the rest here:
CalypsoAI brings testing and validation to ML models used by the TSA - VentureBeat

SPR Announces It Will Give Away $50K Worth of Machine Learning Service to a Local Organization Making a Significant Impact in Its Community – PR Web

October 10, 2022 | Author: admin

"We have seen up close the powerful, positive impact something like machine learning can have on a company. I cant think of a better way to kick off our 50th anniversary than by empowering a local organization to make a significant impact on both its business and community, said CEO Rob Figliulo.

CHICAGO (PRWEB) October 07, 2022

SPR, a technology modernization firm focused on driving exceptional custom technology and digital experiences, announced today that it will be donating $50,000 worth of machine learning (ML) services to an organization that will make a positive impact in the local community. The giveaway, which is open to companies based in either Chicago or Milwaukee, is in honor of SPRs upcoming 50th anniversary.

Based in Chicago and with a significant presence in Milwaukee, SPR is looking to partner with a local company or organization that wouldnt otherwise have access to ML solutions able to take their mission to the next level. Drawing on more than 15 years of data science and engineering experience, as well as its deep expertise in custom software, SPR will be able to provide the winner with services that will allow them to make better-informed decisions more quickly. Depending on their need, SPR can help the winner use its data to predict future outcomes for its organization and services, identify patterns or outliers within their datasets, understand the sentiment of pictures, handwriting, and more, or even use computer vision to detect faces and other objects.

Machine learning is an important emerging technology that can make a huge difference for an organization. Using machine learning, we can train computer systems to perform specific tasks and automate workflows, said Matt Mead, Chief Technology Officer at SPR. This process streamlines projects and offers more accurate results, all while freeing up people to redirect their energy into other important and innovative work. We hope that by donating our time and expertise, the winner of this exciting partnership opportunity will not only advance their organization but also better their community.

SPR has a long history of innovating in and around the Chicagoland and Milwaukee areas. Founded in 1973, SPR has transformed from its humble beginnings as a staffing company into the leading technology modernization firm it is today producing essential custom software and expertly guiding clients through modernization projects.

My father founded this company half of a century ago. Hed be very proud of how far weve come since then, said Rob Figliulo, CEO of SPR. For 50 years, we have been investing in emerging technology, and helping organizations of all types stay ahead of the curve. We have seen up close the powerful, positive impact something like machine learning can have on a company. I cant think of a better way to kick off our 50th anniversary celebrations than by empowering a local organization to make a significant impact on both its business and community.

The giveaway is open to nonprofit and for-profit organizations. All entries will close on November 15, 2022. The official machine learning project will begin in early 2023.

To enter SPRs machine learning giveaway, please visit http://www.spr.com/50k-machine-learning-giveaway/.

About SPRSPR is a digital technology consultancy, focused on delivering exceptional custom technology and digital experiences for the enterprise. We enable companies to do more with data, engage with other people, build disruptive solutions, and operate productively. We're 300+ strategists, developers, designers, and consultants in Chicago and Milwaukee, driving outcomes for our clients through a range of technology capabilities. Our experts know everything from data and cloud implementation to emerging technologies such as blockchain and machine learning. For more information on how SPR delivers beyond the build for its customers, visit http://www.spr.com.

Share article on social media or email:

More:
SPR Announces It Will Give Away $50K Worth of Machine Learning Service to a Local Organization Making a Significant Impact in Its Community - PR Web

Ann Coulter was a hot thirsty mess over homophobe Blake Masters during the Arizona Senate debate – Queerty

October 8, 2022 | Author: admin

Pro-LGBTQ Sen. Mark Kelly and Republican nutjob Blake Masters squared off in a televised debate hosted by Arizona PBS last night and nobody was more parched by it than Ann Coulter.

The 60-year-old homophobe spent a good chunk of the hour-long debate thirsting over 36-year-old Masters, who has never held public office before and is trailing Kelly in the polls.

Lets have a look

It started with Coulter immediately gushing over how very tall Masters looked on stage:

A few minutes later, she declared him absolutely the best person running for Senate this year:

Of course, she couldnt just watch the debate without taking a moment to complain about the ASL translators:

As things were winding down, she begged everyone to watch the hour-long debate online then told them to skip the parts that didnt feature Masters:

Then this morning, she logged onto Twitter to urge peopleIN ALL CAPS!to watch the replay of the debate on C-Span RIGHT NOW!!:

Seriously, girl, go back to your stable and rehydrate yourself at the water trough.

Masters is a venture capitalist who is BFFs with gay billionaire Peter Thiel. Hes also a homophobic pile of garbage who wants to overturn Obergefell v. Hodges, the SCOTUS decision establishing a right to same-sex marriage.

Related: That time anti-LGBTQ candidate Blake Masters went to the gay wedding of his biggest financial backer

According to hisivoterguide profile, Masters also believes religious freedom must be protected at all costs and that businesses should be allowed to discriminate against LGBTQ people if being gay violates their moral and/or religious beliefs.

And on his Twitter page, he has vowed to push for a federal version of Floridas Dont Say Gay law, saying tax dollars should not fund radical gender ideology and weird sex instruction for children.

With views like that, its no wonder he left Coulter totally dehydrated after last nights debate. Voters, on the other hand, seem to be a lot less thirsty for the guy. According to the latestFiveThirtyEight average, Masters is trailing Kelly by six points.

Related: Just when we didnt think Ann Coulter could get any dumber, she did this

Original post:
Ann Coulter was a hot thirsty mess over homophobe Blake Masters during the Arizona Senate debate - Queerty

Researchers use machine learning to assist state food pantries with inventory – Office of Communications and Marketing

October 8, 2022 | Author: admin

Food distribution centers across Alabama have questions. A team of researchers at Auburn University and Tuskegee University are searching for answers.

Which foods and supplies are most needed, where are they most neededand when? Excessive foodsincluding perishablesmight be wasted if supplies are too great. On the other hand, thin supplies could run out quickly, and people depending on this service from their local food pantry might receive nothing.

As the fifth-poorest state in the nation, 17 percent of adults and 23 percent of children in Alabama struggle with food insecurity, according to the Alabama Department of Public Health.

Yin Sun, assistant professor in the Department of Electrical Engineering at Auburn Universitys Samuel Ginn College of Engineering, is part of a multi-disciplinary research project, Strengthening the Alabama Emergency Food Distribution System Using Machine Learning: Impact on Household & Community Food and Nutrition Security During Disasters, across two Alabama universities aimed at solving this problem for the state.

Many distribution centers dont know how many people are coming to collect food, said Sun, who is joined on the project by Tuskegee Universitys Rui Chen, Robert Zabawa, Eunice Bonsi and Souleymane Fall. Sometimes, they have a high number of people come to collect food. Sometimes, they have a low number. This matters greatly when you prepare boxes for people to come pick up. Some of these foods are vegetables, or others, which might require refrigeration or a means of storage to keep the food fresh. These centers constantly deal with this problem.

So many resources, volunteers, time, money and human effort are dedicated to food pantries across the state and nation. Its very important that they serve with the best efficiency possible. That said, we have developed a machine learning algorithm that is trying to predict how many people might come pick up food. This can help the organization from a supply chain and labor resource point of view.

Sun and doctoral students in the universitys Real Time Networking Lab are utilizing data accrued since 2016 from a variety of Alabama food distribution centers in Lee County, including the Food Bank of East Alabama, Lakeview Baptist Church and Auburn United Methodist Church, to develop machine learning algorithms that should help bolster the networks food demand forecasting models, which will vary from location. Sun said food remaining from these distribution centers is often relocated to the Campus Food Pantry at Auburn University.

Developing an algorithm based on factual data will reduce the margin of error that is often a result of guessing based on past experiences, Sun said. We are utilizing historical weekly data and socio-economic data. Its also important to understand how many days schools are open because children are fed from the schools.

This will impact whether parents are coming to pick up food on those days. Then, we must also consider customer personal income. We then weigh these factors into an algorithm that can best create an accurate model.

Sun, who will use what hes learning through this experience to co-teach Applied Statistical and Machine Learning to graduate and undergraduate students at Auburn and Tuskegee in spring 2023, said the team will soon collect data from more food pantries across the stateincluding the states Black Beltto improve the robustness of synthetic data generalization.

Many research topics are regarded as theoretical, Sun said. But this is an area where we have an opportunity to immediately give something beneficial back to society. Ive been working in Auburn for the past five years, and it was very important for me and our students to do something for our community and the state.

Our students are learning cutting-edge research and new technologies. Through this project, they are transferring that knowledge for the greater good.

Read this article:
Researchers use machine learning to assist state food pantries with inventory - Office of Communications and Marketing

Dominos MLops release focuses on GPUs and deep learning, offers multicloud preview – VentureBeat

October 8, 2022 | Author: admin

To further strengthen our commitment to providing industry-leading coverage of data technology, VentureBeat is excited to welcome Andrew Brust and Tony Baer as regular contributors. Watch for their articles in the Data Pipeline.

Domino Data Lab, maker of an end-to-end MLops (machine learning operations) platform, is announcing its latest release version 5.3 today. The delivery includes new support for ML model inferencing on GPU (graphics processing unit) systems and a collection of new connectors. Along with that, the company is beginning a private preview of its Nexus hybrid and multicloud capabilities, first announced in June.

[Also read: Domino Data Lab announces latest MLops platform to satisfy both data science and IT]

GPUs can make lots of ML and deep learning operations go faster because they parallelize massive workloads, which is exactly what training complex deep learning models or numerous ML models entails. For this reason, Domino has long supported GPUs for model training.

But in the case of deep learning specifically, GPUs can benefit inferencing (generating predictions from the trained model) as well, and it is this scenario that Domino newly supports in version 5.3. Perhaps an easier way of thinking about this is that Domino now supports operationalization of deep learning beyond development, extending into production deployment. Given all the new announcements that came out of Nvidias GPU Technology Conference (GTC) last month, Dominos timing here is especially good.

Low-Code/No-Code Summit

Join todays leading executives at the Low-Code/No-Code Summit virtually on November 9. Register for your free pass today.

[See also: Nvidia moves Hopper GPUs for AI into full production]

Then theres the matter of new connectors, including enhanced connectivity for Amazon Web Services S3 and brand new connectors for Teradata and Trino. Usually, new connectors are not newsworthy delivery of them is just a typical, incremental enhancement that most data platforms add at regular intervals. But there are a couple of tidbits here that are worth pointing out.

Coverage of a mature, well-established data warehouse platform like Teradata shows a maturation in MLops itself. Because MLops platforms are new, they often prioritize connectivity to newer data platforms, like Snowflake, for which Domino already had support. But adding a Teradata connector means MLops and Domino are addressing even the most conservative enterprise accounts, where the impact of artificial intelligence (AI) will arguably have the biggest, even if not the earliest, impact. Its good to see the rigor of MLops make its way around all parts of the market.

[Must read: Teradata takes on Snowflake and Databricks with cloud-native platform]

Connecting to Trino an open-source federated query engine derived from Presto development work at Facebook is important in a different way. Connecting to Trino provides further connectivity to all of its target data platforms, including NoSQL databases like MongoDB and Apache Cassandra, data lake standards like Delta Lake and Apache Iceberg, streaming data platforms like Apache Kafka, analytics stores like Apache Druid and ClickHouse, and even productivity data sources like Google Sheets.

[Check out: MongoDB fires up new cloud, on-premises releases]

Finally, theres the Dominos Nexus hybrid/multicloud capabilities, which allow Domino to deploy model training environments across on-premises infrastructure and the three major public clouds, with costing information for each, all from a proverbial single pane of glass. This is pictured in the figure at the top of this post. And because Nexus works across cloud regions, it can also support restricting access to data by geography, to enforce data sovereignty policies and comply with corresponding regulations.

At this time, Nexus is available only to participants in Dominos private preview for same. But progress is progress. Private previews advance to public previews, and public previews eventually progress to general availability (GA). Speaking of GA, Domino 5.3 is generally available now, according to the company. And customers interested can sign up for the Nexus private preview.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.

More here:
Dominos MLops release focuses on GPUs and deep learning, offers multicloud preview - VentureBeat

Filecoin (FIL), Elrond (EGLD), and Flasko (FLSK) are predicted to be the top three investment options for – Bitcoinist

October 8, 2022 | Author: admin

Cryptocurrency is a kind of digital currency that is transferred digitally on the blockchain in a decentralized system to ensure security of transactions. Cryptos may guarantee financial stability because of their independence from centralized bodies such as governments, and banks.

The capacity to execute speedy, low-cost transactions and the durability of decentralized networks that are not prone to a single point of failure are just two of the several benefits that cryptocurrencies offer over their more conventional alternatives. While there are thousands of cryptocurrencies on the market, experts predict that Filecoin (FIL), Elrond (EGLD), and Flasko are the best investment options for 2023.

Filecoin (FIL) is the networks native cryptocurrency. Filecoin (FIL), a cryptocurrency built on the blockchain, claims to provide a decentralized data storage solution. The decentralized structure of Filecoin (FIL) makes it impossible to censor and easy to retrieve data, maintaining its integrity. Filecoin (FIL) allows users to be the gatekeepers of their data while simultaneously expanding internet access across the world.

The incentive of a block reward for data mining and storage on the Filecoin (FIL) network motivates participants to maintain more data and act in an honest manner. Filecoin (FIL) is being traded at around $5.5 which is nowhere near its expected price, which is one of the major reasons Filecoin (FIL) investors are dumping Filecoin (FIL) for Flasko.

Elrond (EGLD) has been affected by the downturn that has plagued the crypto industry since the beginning of this month. Bitcoin (BTC), the most popular cryptocurrency, has a strong positive correlation with Elrond (EGLD).

According to data provided by CoinMarketCap,Elrond (EGLD) has dropped by double digits after a good run in July pushed its value up by 16%. Elrond has been joined by hundreds of fascinating enterprises with the ambition to revolutionize business solutions and create a new online economy. Due to its potential importance in the development of future blockchain-based applications, Elrond (EGLD) might be an important addition to your investment portfolio. There is little hope for Elrond (EGLD) until sometime in the middle of 2023.

Flasko is developing an alternative trading platform for its investors. You may invest in a piece or all of the NFT backed by rare, special, and vintage bottles of whiskeys, champagnes, and wines using Flaskos innovative platform.

Investors in cryptocurrencies may now take part in the rapidly growing alternative investment market, which is now valued at $13.4 trillion thanks to this innovative concept.

Solid Proof has already audited Flasko as a reliable platform and the presale token which was available previously for only $0.015, now costs $0.05, indicating that the Flasko token will increase 5,000% in its price in February 2023. Click on the links below to learn more.

Website: https://flasko.ioPresale: https://presale.flasko.ioTelegram: https://t.me/flaskoioTwitter: https://twitter.com/flasko_io

Disclaimer:This is a paid release. The statements, views and opinions expressed in this column are solely those of the content provider and do not necessarily represent those of Bitcoinist. Bitcoinist does not guarantee the accuracy or timeliness of information available in such content. Do your research and invest at your own risk.

Read the original here:

Filecoin (FIL), Elrond (EGLD), and Flasko (FLSK) are predicted to be the top three investment options for - Bitcoinist

Cloud Hosting

Expert Insights: How to Protect Sensitive Machine-Learning Training Data Without Borking It – DARKReading

AI Can Offer Insight into Who Responds to Anti-Depressants – IEEE Spectrum

Clustering of trauma patients based on longitudinal data and the application of machine learning to predict recovery | Scientific Reports – Nature.com

Fast and scalable search of whole-slide images via self-supervised deep learning – Nature.com

CalypsoAI brings testing and validation to ML models used by the TSA – VentureBeat

SPR Announces It Will Give Away $50K Worth of Machine Learning Service to a Local Organization Making a Significant Impact in Its Community – PR Web

Ann Coulter was a hot thirsty mess over homophobe Blake Masters during the Arizona Senate debate – Queerty

Researchers use machine learning to assist state food pantries with inventory – Office of Communications and Marketing

Dominos MLops release focuses on GPUs and deep learning, offers multicloud preview – VentureBeat

Filecoin (FIL), Elrond (EGLD), and Flasko (FLSK) are predicted to be the top three investment options for – Bitcoinist

Recent Posts

Categories

Archives

Media Sites

Pages

Site admin