Infosys Cobalt Announces Applied AI Cloud Built On NVIDIA DGX A100s

By

-

January 22, 2021

Infosys-NVIDIA-DGX-A100 — Source: NVIDIA

Infosys has made a spectacular entrance into the global cloud-as-a-service arena by offering the Cobalt Applied AI Cloud, built on the latest NVIDIA DGX A100s. The company aims to make AI accessible within the workforce to drive AI-driven transformations for enterprises.

NVIDIA DGX™ A100 systems offer the necessary compute power for enterprise-level AI applications. The A100 systems are considered ‘Workstation in a box.’ They host 1 trillion transistors and offer 5Petaflops clock speed upfront. On top of it, NVIDIA’s inbuilt Multi-Instance GPU (MIG) technology further improves infrastructure efficiency and maximizes the utilization of each DGX A100 system. The systems will provide the required hardware and software stacks needed for over 100 project teams to simultaneously run machine learning and deep learning operations.

Charlie Boyle, Vice President and General Manager of DGX Systems at NVIDIA, said, “Many organizations are eager to infuse their business with AI but lack the strategic platform on which they can pool expertise and scale the computing resources needed to build mission-critical AI applications. Working with Infosys, we’re helping organizations everywhere build their own AI centers of excellence, powered by NVIDIA DGX A100 and NVIDIA DGX POD infrastructure to speed the ROI of AI investments.”

Also Read: Fractal Acquires Zerogons For Its Drag Drop Enterprise AI Platform

The developers and project teams of Infosys can access NVIDIA’s full-stack AI solutions to build services for enterprises. Teams can now develop AI services centrally or locally on any device without any lag. These services will deliver AI-first business processes for enterprises across private and public clouds. Businesses can use these Cobalt AI Cloud services to harness their data and curate data exchanges on the cloud to build and train their AI models. The AI cloud also offers the flexibility to build on top of with services delivered by any hyper-scale cloud provider, to scale and future-proof AI-powered transformation.

Balakrishna D.R., Senior Vice-president of Infosys’ AI & Automation Services, said, “For a long time now, AI has been playing a key role in shaping consumer experience. Cloud, data analytics, and AI are now converging to bring the opportunity for enterprises not just to drive consumer experience but reimagine processes and capabilities too. Infosys applied AI cloud, powered by NVIDIA DGX A100 systems, can help enterprises to quickly build on the opportunity while scaling with new technological advancements.”

Thanks to its recognition in NVIDIA PARTNER NETWORK, Infosys will garner the technical know-how to build NVIDIA DGX A100-powered applications. The premier ones are pre-built ready-to-deploy AI platforms and on-prem AI clouds for enterprises. Other offerings include licensed and open-source AI software-as-a-service (SaaS), models, and edge capabilities.

MeITY And AWS Sets Up Quantum Computing Application Lab

By

Pradyumna Sahoo

-

January 22, 2021

India’s Ministry of Electronics and Information Technology (MeitY) has recently announced that it will set-up a quantum computing application lab with AWS. With this step, India is trying to catch up with the global players like the US, China, and EU in the High-Risk, High-Reward technology.

Quantum computing uses the quantum mechanical properties of matter as the fundamental unit of computation. It has opened new doors in computing, communications, and cybersecurity with wide-spread applications. Google’s quantum supremacy experiment, Quantum encrypted Satellite communication by China, and more have put the nascent field in the spotlight.

India has made a smart and timely move to give its academia and industries access to cutting-edge Quantum computing technology via AWS Braket service, the cloud-based quantum computing service. AWS Braket seems to be the perfect solution for India’s Quantum woes – low quantum hardware and software capacity. The proposed plan states that a steering committee would be set-up to oversee applications around quantum technologies. Primarily, Call for Proposals (CfPs) will be announced from the research community in February-early March. Selected proposals will get AWS Braket credits with access to quantum computing hardware, simulators, and programming tools.

Also Read: OpenMined, In Collaboration With PyTorch, Introduces A Free Course Of “The Privacy AI Series”

During the announcement, the secretary of MeitY Ajay Sawhney said, “The area of quantum computing is at a very nascent stage in the country. The Lab is a great opportunity for our researchers, whether in educational institutions, in research labs, or the startup environment. With quantum computing, there is still a tremendous amount that all of us have to learn together. What exactly will it be useful for, how does it get applied to the scenario that we have within the country? These are the questions we must ask.”

The Indian government is aware of the technological shortfall, especially the lack of hardware and software stack to replicate the domestic software industry’s ecosystem and pose as a global academia leader. It had earmarked 8000 crores rupees for the next five years under its National Mission on Quantum Technologies & Applications (NM-QTA) to develop indigenous quantum technologies.

“The Quantum Computing Application Lab in collaboration with AWS is the first lab on AWS that’s aligned to a government mission. We are trying to pull all of the resources needed to build the capacity and the skills within India to propagate this technology in a meaningful way. The company wants to democratize quantum computing and provide the right development environments to remove the heavy-lifting away from researchers and developers working on quantum algorithms. We want to build this in a scalable manner that can be accessible to all,” said Rahul Sharma, the President of Public Sector – AISPL AWS India and South Asia.

Automated Vehicles’ Safety Standards Relaxed By The US Government

By

Pradyumna Sahoo

-

January 21, 2021

Automated Vehicles Safety Standards — Source: Unsplash

The U.S. government recently exempted automated vehicles (AVs) from some crash standards. The main reason is to cut production costs, promote faster deployment, and amplify mass-adoption. Based on an estimated production of 5.8 million vehicles, the National Highway Traffic Safety Administration (NHTSA) estimates, AV manufacturers would save up to $5.8 billion in the year 2050, i.e., $995 per vehicle.

Previously, the test procedures for AVs were similar to conventional vehicles — the same 80+ tests. Now, the relaxed norms include eliminating the driver’s seat of self-driving vehicles. There will be no crashworthiness standards for automated delivery vehicles as they will never have a human occupant inside of them. These vehicles would still be tested for pedestrian safety and safety-assist systems, but they would no longer need airbags, seat belts, and seats.

For shared passenger-carrying AVs, all front-faced seating positions will now be considered “front outboard passenger seats.” These seats will be subject to traditional regulations, including the placement of advanced airbags. AVs will no longer have a defined driver’s seat, but a child could sit in the front left seat. A vehicle that would permit this, there would be ways to deactivate the baby seat’s airbag.

Also Read: Uber AI Says You Can Increase Task Completion If You Are Polite With Virtual Agents

People have lost lives because there lies confusion between what autonomous vehicles and automated vehicles are. The National Transportation Safety Board stated that no self-driving cars (SAE level 3+) were available for consumers to purchase in 2020. NHTSA had adopted SAE AV classification for Automated Driving Systems which consists of 6 levels – Level 0, Level 1 (“hands-on”), Level 2 (“hands-off”), Level 3 (“eyes off”), Level 4 (“mind off”) and Level 5 (“steering wheel optional”). From SAE 3 onwards, the human driver need not monitor the environment, and the automated system takes over. At SAE 3, the human driver still has a responsibility to intervene when asked to do so by the automated system. The human driver is always relieved of that responsibility at SAE 4, and the automated system will never need to ask for intervention in SAE5.

NHTSA runs the Automated Vehicle Transparency and Engagement for Safe Testing (AV-TEST), which provides an online, public-facing forum for sharing automated driving system on-road testing activities. Automated Vehicle-TEST included more participants after the Partners for Automated Vehicle Education (PAutomated VehicleE) poll suggesting that in American majority do not think autonomous cars are “ready for primetime.”

The test remains controversial because NHTSA relies on voluntary industry interactions. At least 80 companies are testing autonomous vehicles and only 20 have submitted safety assessments to the Transport Department under the current voluntary guidelines set for four years. At the same time, the National Transportation Safety Board (NTSB) has investigated six AV-related crashes revealing problems like inadequate countermeasures to ensure driver engagement, reliance on voluntary reporting, lack of standards, low corporate safety culture, and a misguided oversight approach by NHTSA.

The incompleteness of the data also plagues the test because companies like Pony.ai, Baidu, Tesla, Amazon, Postmates, and Motion have declined, in the past, to provide tracking data. The Automated Vehicles 4.0 (AV 4.0) guidelines also do not mandate regular self-driving vehicle safety assessments. It permits automakers to carry out AV assessments themselves rather than by standards bodies. All these facts simply make the automated vehicle business insidious for the common man. At the same time, it also raises questions about the NHTSA’s credibility to measure safety when the tracking data is not available in the first place.

NASA Uses AI to Detect Martian Craters

By

Pradyumna Sahoo

-

January 20, 2021

The space scientists at NASA’s Jet Propulsion Lab (JPL) have trained an AI that detects new craters from the image data archives of Mars Reconnaissance Orbiter. And as of now, the team has submitted 20+ craters for additional validation.

NASA had deployed its Mars Reconnaissance Orbiter (MRO) on Mars in 2006. It has been 15 years since then, and the orbiter completed 60,000+ orbits beaming back 404 Terabits of data. The two onboard cameras – HiRise and Context have the lion share of the transferred data in the form of high-resolution images.

A crater is produced by the impact of a meteorite, volcanic activity, or an explosion. They are the subject of interest for cosmologists because these structures provide cosmological cues about the age of the planet Mars and the activities going below its surface. The smallest impact-craters are harder to find because their diameter can be as little as 4 meters and to gauge out one such crater from 316 km above the Martian surface requires substantial technical effort. First, the Context Camera detects them in its low-resolution camera covering hundreds of miles at a time. Next, the high-resolution HiRise Camera is used to validate the finding.

Also Read: Optical Chips Paves The Way For Faster Machine Learning

It takes NASA scientists around 40 minutes to manually crawl through one image data to discover new craters. Therefore, JPL researchers trained an AI model that automated fresh impact crater classification to reduce the inference period, from detection to validation. For the model training, the researchers fed 6,830 positive Context Camera images marked with locations of previously discovered impacts confirmed via HiRISE. The training data also had negative examples, images with no fresh impacts. For testing the AI, the researchers used 112,000 Context images. As an immediate result, the whole inference time was reduced from 40mins to an average of five seconds. The NASA AI has detected 20+ craters as of now, and those findings have been submitted for validation via HiRise imaging.

Find more here.

Cure.Fit Goes Global, Acquires US-Based Onyx

By

Pradyumna Sahoo

-

January 19, 2021

The Bangalore-based cure.fit, a health and fitness company, quietly acquired a US-based fitness startup, Onyx, for an undisclosed amount. Cure.fit aims to integrate Onyx’s body tracking technology to provide personalized coaching in its services like live.fit, care.fit, mind.fit, and more.

Cure.fit has secured over $400 million funding across nine rounds recently and is busy spending the money on capacity building. Onyx is the company’s seventh acquisition after acquiring Rejoov, Fitness First India, Seraniti, Kristys Kitchen, a1000yoga, among others.

The global digital fitness market may reach $59.2 billion by 2027, growing at a CAGR of 33.1 percent. Indian market is also likely to reach $2.15 million in 2021 and grow at a CAGR of 2.7 percent to become worth $2.33 billion by 2024, as per Statista. During this time, Cure.fit is undoubtedly making a statement by acquiring an off-shore startup that points towards their international business aspirations.

Also Read: Fractal Acquires Zerogons For Its Drag Drop Enterprise AI Platform

The acquisition will also strengthen the cure.fit’s at-home fitness solutions that currently not at par with their live-training sessions offered at their centers. Hence, cure.fit will use Onyx’s AI platform, which provides human-like personalized guidance in various fields, from physical and mental wellness to nutrition. The platform tracks motion without wearables, eliminating the need for visiting gyms/centers. Onyx’s AI platform also gives trainers the ability to coach a larger audience without physically interacting with them.

“The 20s will be the decade of digital health. Onyx will accelerate our efforts towards building a hardware-agnostic AI-led platform that offers guided content on physical and mental wellness and nutrition all at the same place. Users will get a personalized experience with high-quality tech and human touch, and will be able to achieve their fitness goals from the comfort of their homes without spending on expensive hardware,” Mukesh Bansal, co-founder at Cure.fit said.

Cure.fit currently uses its energy-meter tech, where the phone camera tracks users’ movements as they try to follow guided content from trainers. Adding Onyx’s real-time analysis, cure.fit can improve the personalized measurement element of its platform. The platform can also judge users’ performance, correct issues with the form, and offer properly timed motivations, just like a human coach. This will allow the company to expand the range of tracked activities to include dance and yoga and fast-paced activities like high-intensity interval training while accurately gauging a user’s compliance with proper forms and pacing.

Onyx claims it has “the world’s first 3D motion capture system on your phone” and says it’s capable of tracking “nearly any exercise,” from squats to planks, push-ups, kicks, running, and jumping. A user can place a smartphone in a location capable of capturing the full body during exercise; the app then uses computer vision to render the user as a silhouette while automatically comparing limb and torso motions against ideal representations.

“High accuracy body tracking combined with studio-quality content, will help us create a very differentiated experience for our users,” said Asaf Avidan Antonir, Co-founder of Onyx.

Fractal Acquires Zerogons For Its Drag Drop Enterprise AI Platform.

By

Pradyumna Sahoo

-

January 18, 2021

Fractal continues absorbing new players in the market to build its conglomerate. The recent acquisition, Zerogons, for an undisclosed amount seems to be Fractal’s entry into the highly coveted Cloud industry. This acquisition aims to strengthen Fractal’s Cloud AI business and accelerate the ‘data to decisions’ process for its Fortune 500® clients.

Fractal is one of the most prominent players in the global AI ecosystem with products like Qure.ai that assist radiologists in making better diagnostic decisions, Cuddle.ai helps CXOs make better tactical and strategic decisions, Theremin.ai to improve investment decisions, and Eugenie.ai to find anomalies in high-velocity data.

Although Zerogons Software was a new entrant, it managed to gain a foothold in the market. Thanks to its AI platform Streamflux, users can create workflows using a drag and drop interface (Studio), thereby simplifying the whole ML pipeline from data curation to managing multiple machine learning models at scale. This self-service enterprise AI solution allows anyone to create valuable business solutions using machine learning algorithms without in-depth knowledge. The platform enables the entire lifecycle from raw data to actionable business insights real quick.

Also Read: Cognizant Acquires Inawisdom To Enhance Its AI & ML Capabilities

Pranay Agrawal, the Co-founder and Chief Executive Officer of Fractal, said, “Sandeep and Divya [co-founders] have built a world-class AI/ML engineering platform, and the team at Zerogons. Cloud is the home of AI, and we are excited to partner with Zerogons to operationalize AI for our clients through cloud and engineering.” The Chief Operating Officer of Fractal, Ajoy Singh, said, “We are delighted to welcome the capable team at Zerogons to Fractal. This acquisition strengthens our Cloud Engineering and AI offerings and helps us deliver high velocity and value in our continuing journey to power every human decision in the enterprise.”

Sandeep Mehta and Divya Rakesh were found saying, “We are happy to become a part of one of the leading AI companies in the world. Fractal shares our vision of building the most innovative AI & ML solutions for the industry. With enterprises increasingly demanding cloud-based AI solutions to solve both their immediate and long-term growth objectives, we are well placed as a combined entity to accelerate our clients’ digital journey. We are excited to join a market leader in the AI space. We see tremendous synergies in our joint vision for AI, and together we will help our clients in realizing their short-term and long-term digital transformation objectives”.

All these seem to be an excellent opportunity for both companies to learn and grow together. The integration of Zerogons AI platform with other products shall significantly create new tides in the market, upping the ante against their competition.

Concept-Whitening For Interpreting Neural Networks At Ease

By

Pradyumna Sahoo

-

January 18, 2021

Researchers at Duke University have recently introduced Concept-Whitening, a new type of layer in neural networks that provides the necessary means of interpreting the neural models without hurting predictive performance. The new layer is an alternative to a batch normalization layer as it normalizes and also de-correlates, whitens the latent space — the numerical parameters that store encoded features.

There has been a trade-off between predictive accuracy and interpretability in the machine learning field from the onset of neural networks. The researchers experimented with ConvNets and tested the performance pre-and-post addition of a concept-whitening layer. In ConvNet, the earlier layers detect edges and corners, and the successive layers are built upon those features to detect far more complex attributes. The latent space of the neural model encodes concepts that discriminate classes it is meant to detect. Sadly, the neural models are cheaters. They learn most discriminative features that may not be relevant at all. It is thus vital to know what these models encode in them.

Therefore, a lot of attempts have been made to see inside their hidden layers. In the recent past, there were efforts to interpret individual nodes of pre-trained neural networks. But, the nodes are not always ‘pure,’ i.e., encodes a mixture of features, and information about any concept could be scattered throughout the network. Similarly, Concept-vector — vectors from the latent space chosen to align with predefined or automatically discovered concepts — have also been used. Consequently, they assume each vector encodes only one concept, which is not valid. Hence, these post-hoc approaches rely on the latent space to possess properties that it may not have and can produce misleading and unusable interpretations. Thus, Concept-Whitening emerges as a significant development in deep learning that is featured in Nature Machine Intelligence.

The concepts need not be the labels in the classification problem like the points on any axis that are easier to detect and interpret. The Concept-Whitening module imposes the latent space aligned along the target concepts’ axis. Thus, each point in the latent space has an interpretation in terms of known concepts. The module uses Whitening, which decorrelates and normalizes each axis, along with a rotation matrix that preserves whitening transformation and aligns the concepts with the axes to disentangle concepts.

Also Read: VOneNets – Computer Vision meets Primate Vision

The researchers were quickly able to show a small modification, adding a Concept-Whitening module to neural network architecture, easily visualizing how the network is learning all of the different concepts at any chosen layer. They even showed how concepts are represented at a given layer of the network. The module provides all these perks without hurting predictive performance.

Their experiment with ConvNets revealed that complex concepts are filtered out. The lower layers of the model create lower-level abstract concepts. For instance, an airplane at an early layer is represented by an abstract concept defined by white or grey objects on a blue background. A bed is represented by an abstract concept that seems to be characterized by warm colors (orange, yellow).

In that sense, the Concept-Whitening layer discovers new, more straightforward concepts that can be formally defined and built on, if desired.

Read more here and tinker with the code here.

Amazon Introduces ‘Alexa Custom Assistant,’ Giving Advanced Capabilities To Developers

By

Ratan Kumar

-

January 17, 2021

Amazon debuts’ Alexa Custom Assistant’ to allow companies to build custom assistants. Fiat becomes the first company to leverage the new solution in vehicles. Over the years, building virtual assistants have been a resource-intensive task for organizations as they had to build new skills from the ground up. However, with Alexa Custom Assistant, companies can develop specific skills while using already present Amazon’s existing advanced technologies for Alexa.

Since Alexa Custom Assistant works on top of advanced Alexa technologies — Alexa Skills Kit (ASK), companies will quickly develop and bring products to the market. Organizations now make assistants that can have unique wake word, voice, skills, and capabilities, providing personalized user experience.

Our customers expect to easily connect with their digital lifestyles wherever they go and today we responded with plans to offer new intelligent experiences built on Alexa’s world-class voice AI technology,” said Mark Stewart, Chief Operating Officer, FCA – North America. “We look forward to the expanding partnership with Amazon and the integration of Alexa Custom Assistant within our powerful Uconnect system as we continue on our path to put customer needs and expectations at the center of everything we do.”

Also Read: Google’s Trillion-Parameters NLP Model

What makes Alexa Custom Assistant even more powerful is that Alexa and brand-specific virtual assistants will work hand in hand to deliver a superior customer experience. Both Alexa and bespoke virtual assistants will work simultaneously, improving the skills of virtual assistants. Based on users’ queries, the request will be channeled to either Alexa or brands’ virtual assistants. Such advancement can rapidly increase virtual assistants’ adoption to simplify the way we do our day-to-day tasks.

Alexa Custom Assistants is available at places where Alexa is supported, like U.S, Canada, India, U.K, France, Germany, Japan, Italy, Australia, Brazil, Mexico, and more, but you will have to join the interest list.

Google’s Trillion-Parameters NLP Model

By

Pradyumna Sahoo

-

January 16, 2021

Google loves to scale up things. And no wonder, the trillion parameter threshold for pre-trained language models has been breached by Google AI research. The researchers have recently introduced Switch-Transformers that have 1.6-trillion-parameters. It is the largest sparse-model ever trained for the first time with lower precision (bfloat16) formats.

The basic intuition at work here is to use simple architectures that surpass far more complicated algorithms backed by large datasets and parameter counts. The researchers built the Switch-Transformers’ base architecture on Google’s T5 architectures. They reported a four-fold speed over the T5-XXL and seven times over T5-Base and T5-Large in pre-training speed with the same computational resources.

However, the Switch Transformer fundamentally differs from the currently famous Pre-trained Language Models (PLMs) that use densely activated transformer architectures like GPT-2 and GPT-3. The transformer does not re-use the same weights for all input; instead, it contains a mixture of experts, small models that select different parameters for each input, specialized in various tasks. A gating network looks upon this mixture and draws an inference from the most relevant expert for the task at hand.

This arrangement results in a sparsely-activated expert model with an outrageous number of parameters but provides greater computational efficiency. The sparsity comes from activating a subset of the neural network weights — only the expert model’s weight for each input. The reported computational efficiency was observed from the fact that the 1.6-trillion-parameter model with 2,048 experts (Switch-C) exhibited “no training instability at all,” in contrast to a smaller model (Switch-XXL) containing 395 billion parameters and 64 experts. The researchers credit the efficient combination of data, model, and expert-parallelism to create models with up to a trillion parameters.

This sparse network is distilled to a specialized fine-tuned dense version for a particular downstream task. The researchers were able to reduce the model size by up to 99% while preserving 30% of the sizable sparse teacher’s quality gains. But because of the vast size of the model, novel pre-training and fine-tuning techniques were employed. They came up with a selective precision training that enables training with lower bfloat16 precision. They used a new initialization scheme that enables scaling to many experts and, lastly, increased expert regularization that improves sparse model fine-tuning and multi-task training.

The team trained the huge model by spreading the weights over specialized hardware, TPUs. This training scheme provided a manageable memory and computational footprint on each device. They used the Colossal Clean Crawled Corpus, a 750GB-sized web-crawled multilingual data dataset of text. Masked-training was leveraged where the model had to predict for masked words.

Astonishingly, the model showed a universal improvement across 101 languages and with 91% of languages benefiting from 4x+ speedups over the mT5 baseline. The researchers have a plan to apply the Switch Transformer to multimodal models.

VOneNets – Computer Vision meets Primate Vision

By

Pradyumna Sahoo

-

January 15, 2021

The neuroscientists’ group at MIT-IBM Watson AI lab released VOneNets, a biologically inspired neural network fortified against adversarial attacks. VOneNets are ordinary convolutional neural networks that are more robust by simply adding a new layer, VOneBlock, that mimics the earliest stage of the brain’s visual processing system, V1.

What Is V1?

V1 is the brain’s primary visual cortex that processes visual input like static images and moving ones to recognize edges. And later, the neurons build upon the edges up to the full visual representation of the input. This behavior has inspired ConvNets that detect edges from raw pixels in the first layer, then use the nonlinear combination of edges to detect simple shapes in the next layer. These shapes are again combined in a non-linear fashion to detect higher-level features in subsequent layers. Yet, the ConvNets struggle to recognize objects in corrupted images that are easily recognized by humans.

Scientists thus carried out functional modeling of V1 to emulate the brain’s prowess for visual understanding in computers. As a result, a classical neuroscientific model, the linear-nonlinear-Poisson (LNP) model, was developed. The LNP model consists of a biologically-constrained Gabor filter bank, simple and complex cell nonlinearities, and a neuronal stochasticity generator. And this LNP model became the base of the VOneNets that are superior to ordinary ConvNets.

Also read: Optical Chips Paves The Way For Faster Machine Learning

Why VOneNets?

Can V1 inspired computations provide adversarial robustness if used as a base for initial layers?

Yes, Li et al. had shown in an experiment that biasing a neural network’s representations towards those of the mouse’ V1 increases the robustness of grey-scale CIFAR trained neural networks to both noise and white box adversarial attacks.

How are the activations in ConvNets and primate V1’s related in the context of adversarial robustness?

Using the “BrainScore” metric that compares activations in deep neural networks and neural responses in the brain, the scientists measured all ConvNet model’s robustness. They tested it against white-box adversarial attacks, where an attacker has full knowledge of the structure and parameters of the target neural networks. They found that the more brain-like a model was, the more robust the system was against adversarial attacks.

How does VOneBlock simulate V1?

The LNP model of V1 consists of three consecutive processing stages — convolution, nonlinearity, and stochasticity generator — with two distinct neuronal types, simple and complex cells — each with a certain number of units per spatial location.

The VOneBlock contains elements that mimic the LNP model. A fixed-weight biologically-constrained Gabor filter bank carries the convolutions in VOneBlock to approximate the diversity of primate V1 receptive fields like color and edges. A nonlinear layer introduces nonlinearities with two different nonlinearities, ReLU for simple cell activations and spectral power of a quadrature phase-pair for complex cell activations. And the stochastic layer mimics stochastic neural behavior — repeated measurements of a neuron in response to nominally identical visual inputs resulting in different spike trains.

How do VOneBlock is integrated into VOneNets?

The VOneNet replaces the first stack of convolutional, normalization, nonlinear, and pooling layers in a ConvNet with the VOneBlock and a trained transition layer.

How do VOneNets perform against adversarial attacks?

VOneNets are substantially more robust than their corresponding base models and retain high ImageNet performance after training. The adversarial robustness permeates across all architectures, hence shows the generalisability of the VOneblocks. All properties of the VOneBlock — Convolution, nonlinearities, and stochasticity — work in synergy to improve robustness to different perturbation types. And the neuronal stochastic elements at the VOneBlock level lead the downstream layers to learn more robust input features.

The blue region represents the increment in white-box accuracy over the base models.

Read about the research in detail here and find the code here.