Solving the scaling errors in Optical Neural Networks

December 11, 2022

Neural network has been serving as the backbone of nearly every notable achievement in deep learning-based AI technologies. Artificial neural networks, a type of advanced deep learning algorithm, have drawn a lot of interest for their potential use in fundamental tasks, including language processing and image recognition. However, the rapidly rising energy costs associated with ever-larger neural networks and higher processing demands are a barrier to further advancement. Optical neural networks have the ability to alleviate the energy cost and computational issues that other models suffer. These deep learning architectures operate multiple times faster and use much less energy by relying on light signals rather than electrical impulses.

The core idea of artificial neural networks is based on the computational network models present in the nervous system. Several artificial neural network approaches, such as convolutional neural networks and recurrent neural networks, employ matrix multiplications and nonlinear activations (the functions that mimic how neurons in the human brain respond). The functionality and interconnectedness of neurons can be implemented in optical neural networks by using optical and photonic devices and the nature of light propagation. While nonlinear activation functions are normally implemented by either the optoelectronic method or the all-optical method in optical neural networks, optical components are frequently employed for linear functions. This is because nonlinear optics typically calls for high-power lasers, which are challenging to implement in an optical neural network.

In optical analog circuits, its linear unit multiplies an input vector and a weight matrix. One of them is a circuit that can implement a certain class of unitary matrices with a constrained number of programmable Mach-Zehnder interferometers (MZIs) as its weight matrix. A Mach-Zehnder interferometer is a type of connected, reconﬁgurable, adjustable mirrors which constitutes an optical neural network. A typical MZI has two beam splitters and two mirrors. The top of an MZI receives light, which is split into two pieces that interfere with one another before being recombined by the second beam splitter and reflecting out the bottom to the following MZI in the array. Researchers can process data by performing matrix multiplication using the interference of these optical signals. The circuit does a good job of balancing the performance of the ONN with the number of programmable MZIs. As a result, optical neural networks built on a set of cascading MZIs are being considered as a potential alternative to current deep learning technology.

When compared to their electronic equivalents, optical network-based devices may provide superior energy efficiency and processing speed. One can modify each MZI’s output to facilitate the imitation of any matrix-vector multiplication by using programmable phase shifters. The programmability of ONNs depends on these phase shifters, but on the other hand, learning the MZI parameters of the circuit with a traditional automated differentiation (AD), which machine learning platforms are equipped with, takes a lot of time.

In addition, errors that might arise in each MZI soon compound as light passes from one device to the next. There are situations where it is difficult to tune a device such that all light flows out the bottom port to the next MZI due to the fundamental design of an MZI. If the array is very vast and the device loses a small amount of light at each stage, there will only be a very small amount of power remaining in the end. As a result, it is impossible to program the MZI to the cross-state perfectly. This results in component errors, which prevent programmable coherent photonic circuits from scaling.

Some errors can be avoided by anticipating them and configuring the MZIs such that subsequent devices in the array will cancel out earlier errors. Several studies have focused on “correcting” hardware errors by global optimization, self-configuration, or local correction. Even though correction decreases mistakes for standard MZI meshes by a quadratic factor, not all errors get eliminated. Error effects continue to develop with mesh size, posing a fundamental constraint to the scalability of these circuits.

Recently, a group of MIT researchers suggested two mesh architectures that accomplish the same perfect scaling: a 3-splitter MZI that corrects all hardware errors and an MZI+crossing design. Instead of the usual two-beam splitters, 3-MZI has three. The extra beam splitter combines the light, making it considerably easier for an MZI to get the necessary setting to send all light from its bottom port. The team notes that because the additional beam splitter is a passive component and only a few microns in size, it doesn’t require any more wiring and doesn’t significantly alter the size of the chip.

The researchers discovered that their 3-MZI architecture could substantially minimize the uncorrectable arbitrary error that affects accuracy when they tested it using simulations. The amount of error in the device actually decreases as the optical neural network grows larger, which is the reverse of what happens with a device using conventional MZIs. With an error that has been decreased by a factor of 20, researchers could build a device large enough for commercial usage using 3-MZIs. The MIT team demonstrated that this improved MZI mesh is >3x more resilient to hardware errors using a benchmark optical neural network, enabling effective inference in a regime where conventional interferometric circuits fail.

The MZI+crossing architecture corrects correlated errors and has the added benefit of having a larger intrinsic bandwidth, which allows the optical neural network to run three times faster. The correlated errors are caused by manufacturing flaws; for example, if a chip’s thickness is slightly off, the MZIs may all be off by around the same amount, and the faults will thus be roughly the same. To make an MZI more resilient to these kinds of faults, MIT researchers tried to modify its configuration through this design.

In addition to requiring no extra phase shifters, this design uses a lot less chip space than the ideal redundant MZIs. The proposed architecture designs also offer progressive self-configuration, enabling error correction even when the source of the hardware errors is unknown. This research will pave the way for the creation of freely scalable, broadband, and compact linear photonic circuits.

The MIT researchers intend to test these architecture techniques on actual hardware now that they have demonstrated these techniques using simulations, and they will keep working toward an optical neural network they can successfully implement in the real world.

The U.S. Air Force Office of Scientific Research and a graduate research scholarship from the National Science Foundation both contributed to the funding of this study.

The study, which was published in Nature Communications, was led by Ryan Hamerly, a senior scientist at NTT Research and a visiting scientist at the MIT Research Laboratory for Electronics (RLE) and Quantum Photonics Laboratory. The paper was co-authored by graduate student Saumil Bandyopadhyay and senior author Dirk Englund, an associate professor in the MIT Department of Electrical Engineering and Computer Science (EECS), the leader of the Quantum Photonics Laboratory, and a member of the RLE.

Microsoft India and LinkedIn announce the Skill for Jobs program

Manjiri Gaikwad

December 11, 2022

Two leading companies, LinkedIn and Microsoft India, announced the Skill for Jobs program, which provides access to 350 courses and six new career essential certificates to get six of the highest in-demand jobs in the digital economy.

Both tech companies will provide 50,000 LinkedIn learning scholarships to help people in India to enhance their skills. They will train and certify 10 million people with appropriate skills for in-demand jobs by the end of the year 2025.

The Skill for Jobs program is built on the Global Skills Initiative, which helped about 80 million job applicants access digital skilling resources worldwide. So far, Microsoft has engaged 40 million learners in Asia with the help of LinkedIn, Microsft Learn, and non-profit skilling practices. Out of these 14 million learners, 7.3 million were from India.

LinkedIn and Microsoft used LinkedIn and Burning Glass Institute data to analyze job listings to detect the programs’ six most in-demand jobs. The six jobs are Administrative Professional, Project Manager, Business Analyst, Software Developer, Data Analyst, and System Administrator. The program will offer courses and certifications in seven languages French, English, German, Portuguese, Japanese, Spanish, and Chinese.

National Technology Officer, Microsoft India, Dr. Rohini Shrivathsa, mentioned that bridging the skills gap in today’s digital economy is fundamental to India’s employment challenges and societal progress. Microsoft has constantly been investing in various initiatives to skill India’s youth, bring out the potential of underserved communities and provide opportunities to empower women into the workforce.

SALIENT: MIT technology improves Training and Inference Performance in Graph Neural Network

Preetipadma K

December 11, 2022

Most people visualize a pair of perpendicular lines describing the relation between two entities using a line, a curve, or bars of various heights when they hear the term “graph.” However, in data science and machine learning, a graph represents a two-part data structure: Vertices and Edges, or G = (V, E). Here, V denotes the collection of vertices, while E denotes the edges connecting the vertices. Today, graphs are being used to represent and examine relationships between data, including those relating to social networking, finances, transportation, energy grids, learning molecular fingerprints, predicting protein interface, and disease classification. Graph neural network is a subfield of machine learning that focuses on creating neural networks for graph data as efficiently as possible. They help data scientists work with data in non-tabular formats.

Building mini-batches in graph neural networks is extremely computationally expensive due to the exponential growth of multi-hop graph neighbors over network layers, in contrast to regular neural networks. This makes it difficult to enhance the training and inference performance of graph neural networks. In order to solve these problems, MIT researchers worked with IBM Research to create a novel approach dubbed SALIENT (SAmpling, sLIcing, and data movemeNT). Their technique drastically reduces the runtime of graph neural networks on large datasets, such as those with 100 million nodes and 1 billion edges, by addressing three primary bottlenecks. The newly developed method also scales well when the computational capacity is increased by one to sixteen graphical processing units.

The research was presented at the Fifth Conference on Machine Learning and Systems. It was supported by the U.S. Air Force Research Laboratory and the U.S. Air Force Artificial Intelligence Accelerator, as well as by the MIT-IBM Watson AI Lab.

The need for SALIENT became even more evident when researchers started investigating at the challenges that current systems faced when scaling cutting-edge machine learning algorithms for graphs to enormous datasets, practically at the order of billions. The majority of current technology achieves adequate performance on smaller datasets that can easily fit into GPU memory.

According to co-author Jie Chen, a senior research scientist at IBM Research and the manager of the MIT-IBM Watson AI Lab, large datasets refer to scales like the whole Bitcoin network, where certain patterns and data links might indicate trends or criminal activity. Of the blockchain, there are around one billion Bitcoin transactions. If researchers wish to spot illegal activity inside such a vast network, they will have to deal with a graph similar to this size. The team’s main objective is to build a system that can manage graphs that may be used to represent the whole Bitcoin network. To keep up with the rate at which new data is created almost daily, they also want the system to be as efficient and streamlined as possible.

The team worked on building SALIENT with a systems-oriented approach that included basic optimization techniques for components that fit into pre-existing machine-learning frameworks, such as PyTorch Geometric (a popular machine-learning library for GNNs) and the deep graph library (DGL), which are interfaces for creating a machine-learning model. The key objective of developing a technique that could easily be included in current GNN architectures was to make it intuitive for domain experts to apply this work to their specialized domains in order to speed up model training and pluck out insights during inference faster. The team modified its architecture by continually using all available hardware, such as GPUs, data lines, and CPUs. For instance, while the CPU samples the graph and prepares mini-batches of data to be sent across the data link, GPU would either train the machine-learning model or perform inference.

The researchers started by examining the performance of PyTorch Geometric, which revealed an astonishingly low usage of the available GPU resources. The researchers increased GPU usage from 10 to 30% by using minor modifications, which led to a 1.4 to 2 times performance increase compared to open-source benchmark codes. With this fast baseline code, the algorithm could run through one full pass (or “epoch”) on a large training dataset in 50.4 seconds. The MIT research team set out to examine the bottlenecks that develop at the beginning of the data pipeline as well as the algorithms for graph sampling and mini-batch preparation since they felt they might get even improved results.

In contrast to conventional neural networks, GNNs carry out a neighborhood aggregation operation, which calculates information about a node using input from other neighboring nodes in the graph, such as information from friends of friends of a user in a social network graph. The number of nodes a GNN must connect to also increases with the number of layers, which might occasionally strain a computer’s capabilities. Although some neighborhood sampling methods employ randomization to boost efficiency marginally, this is insufficient because the methods were developed when contemporary GPUs were still in their infancy.

To overcome this, the team came up with a combination of data structures and algorithmic improvements that increased sampling performance. As a result, the sampling operation alone was enhanced by approximately three times, decreasing the runtime per epoch from 50.4 to 34.6 seconds. The team adds that they uncovered a previously overlooked fact: sampling can be done during inference at an appropriate rate, boosting total energy efficiency and performance.

According to MIT, earlier systems used a multi-process strategy for this sampling stage, resulting in additional data and pointless data transfer between the processes. By building a single process with small threads that retained the data on the CPU in shared memory, the researchers improved the dexterity of their SALIENT technology. The researchers also point out that SALIENT utilizes the shared memory of the CPU core cache to parallelize feature slicing, which collects relevant data from nodes of interest and their immediate surroundings and edges. As a result, the overall runtime per epoch dropped from 34.6 to 27.8 seconds.

The last bottleneck included a prefetching step to streamline the exchange of mini-batch data between the CPU and GPU, which helps prep data right before it’s needed. The researchers predicted that doing so would use all of the available bandwidth on the data link and bring the approach up to 100% utilization, but they only witnessed about 90%. They also discovered and solved a performance issue in a well-known PyTorch library that resulted in redundant round-trip CPU and GPU connections, giving SALIENT a runtime of 16.5 seconds per epoch.

The team believes that they were able to achieve such outstanding outcomes because of their painstaking attention to detail. The team hopes to apply the graph neural network training system to the existing algorithms used to categorize or forecast the features of each node in the future, as well as focus on identifying deeper subgraph patterns. The latter can benefit in identifying financial crimes.

Should Government Employ Sentiment Analysis To Better Understand Its Citizens?

Preetipadma K

December 10, 2022

People often rush to social networking platforms like Facebook, Instagram, and Twitter and other online platforms like LinkedIn, YouTube to share trivial and important events from their lives. Whether it is mundane or exciting, there is an innate urge to talk about current affairs and vapid details about everything. Be it ranting about how they feel about certain things – stuck in a traffic jam, poor customer service, festival season, etc. – people have the compulsion to communicate their thoughts to the world via social media. While this seems pointless, these ‘thoughts’ can be data points for the government to understand its citizens via sentiment analysis.

Sentiment analysis is the practice of using Natural Language Processing (NLP), a subset of machine learning, to recognize sentiments and emotions that are communicated in human languages. NLP enables machine learning-powered systems to recognize and evaluate the thoughts and opinions expressed in a post or any piece of content uploaded on the internet. First, NLP recognizes human language and transforms it into machine language. After the natural language input has been converted into machine-understandable text, natural language processing algorithms scan it to look for trends, anticipate future events, or discern the precise message the user intended to communicate. Leveraging this technology in the business allows organizations to pinpoint if a customer has a favorable or negative opinion of their good or service and take the appropriate action to resolve it.

Businesses get insight and understand exactly what is expected of them so they can respond in real time by analyzing the thousands of comments and opinions made on social networking sites, online surveys, reviews, and sometimes even videos. This has been a driving factor for reliance on sentiment analysis by brands looking to be thoughtful regarding what they are talking about, how they are marketing themselves, and how they are accountable for their actions while safely inclining toward our interests.

Sentiment analysis is used in various fields, including business and marketing, journalism, and others. It has a wide range of applications that can help decision-making across many domains. Even government organizations can employ sentiment analysis to gain better insights into the opinions of their citizens. They can use it during elections, public action or even manage and monitor a natural disaster.

As was noted in the introduction, social networking sites like Twitter, LinkedIn, and others are some of the most popular avenues for people to express their thoughts on a broad range of topics. Microblogging sites like Twitter can be an excellent breeding ground for sentiment analysis owing to the accessibility, visibility, and abundance of Twitter content. Twitter allows people to express themselves by posting and interacting with brief tweets. Every day, millions of tweets are posted on virtually every subject. This makes Twitter a resourceful platform for sentiment analysis. Besides, Twitter is the only platform that still caters to Genz, millennials, and boomers alike.

Meanwhile, platforms like LinkedIn can be viewed as collections of hyperactive communities engaging about issues in a detailed manner. Here, the narratives about any topic are directly controlled by an ‘influencer’ or ‘social celebrity,’ who exerts such influence because of their professional career, large followers, or both. In comparison to Twitter, Linkedin has quite a formal tone.

In smart cities and other cutting-edge urban jungles where microblogging platforms like Twitter and apps like LinkedIn are increasingly widely utilized, the government can employ NLP on such platforms to clearly understand the issues and grievances of their population as they arise via sentiment analysis.

Every government aspires to improve the effectiveness of its public services and take the initiative in order to serve its population better. Sentiment analysis can assist them in comprehending the primary problems that citizens encounter, such as traffic congestion and inconsistent treatment at the government offices and federal agencies like the police. The insights from such analysis can assist them in developing new policies to address those concerns, and grievances and get real-time reactions from people to them. The latter will aid government employees in comprehending how public opinion has been impacted by government initiatives and policies.

Ruling political parties can plan new policies or election campaigns based on the causes of people’s angst or happiness and the associated patterns. Even opposition parties can identify trends in people’s resistance or support for new legislation and develop agendas for their own parties through sentiment analysis. Sentiment analysis enables both sides of government to remain aware of the public opinion about their parties, their actions, and their statements. One incorrect comment may affect public opinion negatively on social media, which can lead to fatal reputation damage during an election season.

It is possible that government may hesitate from leveraging sentiment analysis due to privacy violations. However, that can be addressed by having social media user data protection laws in place that can check how the government is using the data, data storage, data selling and other privacy concerns.

UAE’s new AI-enabled system to handle employment contracts

Manjiri Gaikwad

December 10, 2022

The Ministry of Human resources and Emiratisation (MoHRE) develops an AI-enabled automated system to complete employment contracts without the intervention of humans.

The new AI system is a part of the National Artificial Intelligence Strategy that focuses on setting the UAE as a global leader in artificial intelligence by building an integrated framework in different areas of the UAE.

As per the Ministry, 35000 employment contracts have been completed in the first two days of the new AI system’s launch. These contracts consist of new and renewed employment contracts that were approved after the signatures of both parties. The new system uses advanced technologies to process and verify images in the contracts, reducing the time from two days to just 30 minutes.

Besides AI-enabled systems, UAE has also acquired many technologies like the awareness program via the self-guidance service, the smart mobile app, the smart communication framework, the Nafis platform, and more. Users can access the awareness program via the Ministry’s smart mobile app, which provides over 100 ministry services powered by big data and AI. The WhatsApp channel of the Ministry is the first federal entity to have a verified business account on the Meta-owned app. The channel is available in both English and Arabic languages.

Why is the Singapore government’s workplace AI therapy platform Midline at Work problematic?

Sahil Pawar

December 10, 2022

Singapore government's Midline at Work problematic

The Singapore Ministry of Health and Ministry of Education launched the Mindline at Work tool in August 2022 for public-sector teachers as part of a mental health initiative. A chatbot component was trialed during development. However, anxious and burned-out teachers coming to the portal for help were met with comments like: “Remember, our thoughts are not always helpful. If a family member or friend was in your place, would they also see it the same way?” Unsettling screenshots of the chatbot’s interactions went viral on social media platforms.

Wysa

As part of the mental health initiative, the Singaporean government partnered with Wysa, which is a well-known name in the AI therapy app sector. Wysa is recognized for having one of the most substantial evidence bases among several such apps and is clinically recommended by expert groups such as the Organisation for the Review of Care and Health Apps.

Despite that, in an investigation by Rest of World, users described Mindline at Work as a one-size-fits-all software that struggles to meet the specific needs of the teachers. More generally, psychology experts warn that partnering with digital therapy or wellness apps can backfire when the leading causes of mental health issues, in this case at workplaces, remain unaddressed.

Singapore Government’s Mental Health Initiative

Singapore’s government is the first to bring Wysa’s therapy bot into a national-level service. The original Mindline.sg initiative, which was launched in June 2020, was aimed to help anyone in Singapore access care during the pandemic. The platform was integrated four months later as an “emotionally intelligent” listener by the Ministry of Health’s Office for Healthcare Transformation, Singapore. Later on, when teacher burnout during the pandemic became a prominent news topic, an extension, Midline at Work, was rolled out for public education professionals as a more tailored version.

Midline at Work’s Generic Advice

Complaints began to emerge only a few days after the launch of the extension. Upset users were unsatisfied with the chatbot’s generic advice. They said it is not the right tool to address the root causes of teachers’ stress, including uncapped working hours, demanding performance appraisal systems, and large classroom sizes.

“It’s pretty useless from the teacher’s point of view,” said a public school teacher in his late 20s. He said that no one he knows takes the Midline bot seriously. “It’s a joke. It is trying to gaslight the teachers into saying, ‘Oh, this kind of workload is normal. Let’s see how you can reframe your perspective on this,'” he said.

Experts Take

In response to the backlash, Sarah Baldry, Wysa’s vice president of marketing, said that the app helps its users to build emotional resilience. “It is important to understand that the chatbot can’t change the behavior of others. Wysa can only help users change the way they feel about things themselves.”

Some research shows that AI apps do promise to alleviate symptoms of anxiety and depression, although a significant issue is that most of them are not evidence-based. Other barriers that make AI bot therapy unpopular are concerns about data privacy and low engagement. Being the founder of the mental health collective Your Head Lah, Reetaza Chatterjee works in the nonprofit sector. “I don’t trust that these apps wouldn’t share my confidential information and data with my employers,” she said.

Overall, privacy remains a weak point. Mozilla Foundation, a digital rights nonprofit, found that many mental health apps majorly fail at it. However, Mozilla noted that Wysa was a strong exception in this case, though users who commented on the matter hadn’t used the platform enough to think about it.

Conclusion

Although artificial intelligence is drastically transforming the healthcare industry, particularly the mental health sector, it is evident from the example of Midline at Work that it cannot be completely relied upon, and doing so can prove disadvantageous. AI still has a long way to go before it can genuinely achieve human intelligence. Until then, we can make the most of what the technology has to offer to radicalize mental health scenarios while keeping an eye out for any irregularities and abnormalities.

Deepfake Detection Technology: Where do we stand?

Preetipadma K

December 9, 2022

Deepfake ai detection — Credit: Mihai Surdu/Shutterstock

A team of researchers from the Johannes Kepler Gymnasium and the University of California, Berkley, created an artificial intelligence (AI) application that can determine whether a video clip of a famous person is real or a deepfake.

To determine if a video was real or not, researchers Matyá Boháek and Hany Farid discuss training their AI system to distinguish certain persons’ distinctive physical movements in their study published in Proceedings of the National Academy of Sciences. Their research expands on previous work in which a system was trained to recognize deepfake features and head movements of prominent political figures, including former president Barack Obama.

Deepfakes, a portmanteau of the terms “deep learning” and “fake,” initially appeared on the Internet in late 2017, powered by generative adversarial networks (GANs), an exciting new deep learning technology. Today, deepfakes have plagued the internet with their presence.

Consider the following scenario: you receive a video of a celebrity from a friend. You see that the famous celebrity is making an absurd statement or having a dance-off or engaging in ethically questionable activity. Whether intrigued or shocked, you forward the video to your other friends, only to discover later that it is fake. Now think back to the time you first watched the video. Perhaps you assumed the video was real cause it appeared to look completely like one. Unlike earlier deepfake videos, which were quickly debunked in the previous decade, today, GANs are powerful enough to create deepfake content where the human eye cannot discern if it is manipulated media.

In February, a study published in the Proceedings of the National Academy of Sciences USA, found that humans are finding deepfake images to be more realistic than the actual ones. Researchers at Stanford Internet Observatory reported in March that they had found over 1,000 LinkedIn accounts with profile photos that seemed to be generated by AI. Such instances highlight that it is important to develop a tool or solution that identifies the deepfake content online.

Last month, Intel introduced new software that is claimed to be capable of instantly recognizing deepfake videos. With a 96% accuracy rate and a millisecond reaction time, the company asserts that their “FakeCatcher” real-time deep fake detection is the first of its type in the world.

In their current research, Boháek and Farid trained a computer model using over 8 hours of video footage that shows Ukraine President Volodymyr Zelenskyy saying things he did not say. According to reports, the video was produced to support the Russian government in persuading its public to believe state propaganda about the invasion of Ukraine.

Inspired by the previous research study where AI could identify deepfake by analyzing the jagged edges of pupils of human eye, Boháek and Farid noted at the outset that people have other distinctive qualities aside from physical markings or facial features, one of which was body movements. For instance, they discovered Zelenskyy’s tendency to raise his right eyebrow whenever he lifts his left hand. They used this information to create a deep-learning AI system to analyze a subject’s physical gestures and movements by reviewing video footage of Zelenskyy. Over time, the system became more adept at identifying actions that people are unlikely to notice—actions that collectively were exclusive to the video’s topic. It can recognize when something doesn’t match a person’s regular patterns.

Up to 780 behavioral characteristics are analyzed by the detection system as it examines many 10-second segments obtained from a single video. It will alert human experts to take a closer look if it flags many segments from the same video as being fake.

The researchers then tested their system by evaluating multiple deepfake films together with genuine videos of different people. Researchers obtained true positive rates of 95.0%, 99.0%, and 99.9% when comparing different subsets of an individual (facial, gestural, or vocal) or combination characteristics against various datasets. This suggests that they discovered their system successfully distinguishes between actual and deepfake. It was also successful in recognizing the fabrication of the Zelenskyy video.

Though this is an exciting and comforting news, there is a catch! While the success rates of the deepfake detection tools are encouraging, the presence of misinformation and misleading content will not disappear. As AI becomes adept in recognizing deepfakes, its technologies are also helping in the creation of more powerful deepfakes which can evade the existing technologies. Hence these detection solutions offer a partial solution to countering the threat. However, they do present a fighting chance to minimize the harm caused by deepfake content.

Uber Launches Self-Driving Robotaxis in Las Vegas

Preetipadma K

December 9, 2022

Uber Launches Self-Driving Robotaxis in Las Vegas — Image Credits: Motional

Uber is collaborating with driverless technology startup Motional to provide autonomous car rides in Las Vegas, with hopes to extend to other major cities such as Los Angeles.

Starting on Wednesday, Uber users in Las Vegas may request an autonomous version of the Hyundai-designed IONIQ 5 mid-sized SUV. The Motional modified IONIQ 5 has 30 outside sensors, together with cameras, radar, and lidar systems, which enable Level Four autonomy by identifying threats at an ultra-long range. In other words, automobiles can operate autonomously, although in a few specific situations and only in pre-approved (geofenced) areas.

The other features of the Ioniq 5s remain the same, including a range of 238 to 315 miles and 220kW quick charging, which enables the batteries to refuel from 10% to 80% charge in 18 minutes.

The self-driving cars cannot be requested directly by customers. Instead, users must choose “UberX” or “Uber Comfort Electric” to be paired with a Motional vehicle. Customers will need to opt-in before the trip is confirmed, and a self-driving car will be sent to pick them up if one is available. Two “vehicle operators” are dispatched in the robocars to keep an eye on the technology and offer further assistance to passengers. According to Motional, if a robotaxi encounters challenging circumstances (such as road construction or flooding), an operator can remotely manage it to steer it to safety. Once the robotaxi has arrived at the agreed-upon pickup location, the Uber app will urge users who have accepted it to unlock the doors.

The companies aren’t charging for driverless rides, but they plan to do so once the service is available to the general public.

The debut of the robotaxi service comes after the introduction of self-driving Uber Eats deliveries in Los Angeles in May as part of a 10-year business agreement between Uber and Motional.

Exactly two years after selling its self-driving division to Aurora Innovation, the launch ushers Uber into a new age of autonomous vehicles. Automation has long been considered a means of reducing costs and accelerating service by rideshare companies like Uber and Lyft.

With the potential to serve millions of customers, Uber claims that its recent cooperation with Motional would result in one of the largest deployments of autonomous vehicles on a major ride-hailing network.

Since 2018, Motional has been providing robotaxi services in Las Vegas through Lyft, a competitor of Uber; however, until 2020, trips were provided under the parent company Aptiv.

NeuReality Receives Series A Funding of US$35 Million

Preetipadma K

December 9, 2022

NeuReality, an Israeli AI chip startup, has announced a US$35 million Series A fundraising round to commercialize its NR1 processor, which is developed to speed up artificial intelligence applications. The round was headed by Samsung Ventures, Cardumen Capital, Varana Capital, OurCrowd, and XT Hi-Tech, with participation from SK Hynix, Cleveland Avenue, Korean Investment Partners, StoneBridge, and Glory Ventures. With this round, NeuReality’s total funding now stands at US$48 million.

The NR1, which is a network-attached “server on a chip,” employs a new class of Network Addressable Processing Units (NAPU) designed specifically for deep learning inference applications such as computer vision, natural language processing, and recommendation engines. Large-scale users like Hyperscalers and next-wave data center clients will be able to accommodate the expanding spectrum of their AI usage thanks to the NAPU.

The latest funding will bolster NeuReality’s aspirations to begin implementing its Inference products in 2023. The term “inference” refers to the process of executing trained neural networks in production. In contrast to the existing technologies, NeuReality’s solution is designed for optimum deployment in data centers and near-edge on-premises sites. These locations require better performance, reduced latency, and significantly higher efficiency. Generally, large-scale AI- infrastructure settings struggle with maintaining hardware efficiency with scaling demands. This is because they require the addition of more chips to the infrastructures – which in turn requires huge power to manage them. The NR1 chip solves the problem via linear scaling, where you add more chips to the server cluster without compromising hardware efficiency. At the same time, the latency of AI operations is reduced, and system costs and power usage are reduced. These factors are crucial for improving the total cost of ownership (TCO) of data centers and on-premises large-scale compute systems, which is essential for the business models of many applications.

NeuReality provides the NR1 as part of an appliance called the NR1-S Inference Server, which features several NR1 chips. When compared to rival hardware, NeuReality claims that the NR1-S Inference Server can reduce prices and power needs by a factor of 50. The company also features the NR1 as part of the NR1-M accelerator card, which can be connected to a server via a PCIe port. With the use of the accelerator card, companies can incorporate NeuReality’s technology with their current server infrastructure in their data centers.

In addition to the NR1, NeuReality offers a collection of software tools to make deploying AI applications in production easier. These solutions from the company also promise to make managing applications easier. Among the software components in NeuReality’s portfolio is an AI hypervisor, which assists customers in managing machine learning applications deployed on NR1 chips.

Dr. Mingu Lee, Managing Partner at Cleveland Avenue Technology Investments, said, “NeuReality is bringing ease of use and scalability into the deployment of AI inference solutions, and we see great synergy between their promising technology Fortune 500 enterprises companies we communicate with. We feel that investing in companies such as NeuReality is vital, not only to ensure the future of technology, but also in terms of sustainability.”

Since last May, NeuReality claims it has been distributing NR1 prototype implementations to partners. Using its latest US$35 million funding round, the company hopes to roll out its technologies extensively. In order to help with the endeavor, NeuReality will hire 20 additional staff over the next six months.

GitHub Announces GitHub Copilot for Business Plan

Preetipadma K

December 9, 2022

Last year in June, GitHub and OpenAI debuted a technical preview of a new AI tool called Copilot, which runs within the Visual Studio Code editor and autocompletes code snippets. Now, months after debuting for individual users and schools, GitHub Copilot is now available in a plan for enterprises.

The new subscription, known as GitHub Copilot for Business, costs $19 per user per month and includes all the capabilities of the Copilot single-license tier in addition to flexible licensing management, organization-wide policy controls, and industry-leading privacy. With the help of this plan, businesses can easily establish policy controls to impose user preferences for public code matching on behalf of their company.

The emergence of AI-assisted coding, according to GitHub’s blog post, will transform how we create software, much like the rise of compilers and open source. Because of this, GitHub has faith in the ability of AI to enhance the developer experience, boost productivity and happiness, and speed up innovation by offering GitHub Copilot to businesses of all sizes with enhanced admin controls.

GitHub wants to offer more tools in 2023 to assist developers to make educated judgments about whether to adopt Copilot’s suggestions, such as the ability to recognize strings matching public code and link to those repositories. Additionally, GitHub asserts that it will not save or distribute code samples for training purposes with GitHub Copilot for Business users, regardless of whether the data originates from public repositories, private repositories, non-GitHub repositories, or local files.

This development follows the filing of a class-action lawsuit in a federal court in the US challenging the legality of GitHub Copilot and OpenAI Codex. The lawsuit filed against GitHub, Microsoft, and OpenAI alleges a breach of open-source licensing and has the potential to significantly influence the artificial intelligence community. The lawsuit was filed by Matthew Butterick, a programmer and lawyer, and the legal firm Joseph Saveri, which specializes in antitrust and class lawsuits.

1...119120121...354 Page 120 of 354