Articles tagged with "machine-learning"

Automate And Speed Up TCAD Calibration With Expert Modules And ML Calibration Accelerator

The article discusses the importance of automating and speeding up TCAD calibration in semiconductor manufacturing using expert modules and machine learning (ML) calibration accelerators. It highlights the challenges in semiconductor development due to increased complexity and the need for efficiency. TCAD calibration involves tuning physical model parameters to match simulated results with real device data. Synopsys offers solutions like Sentaurus Calibration Workbench (SCW) with expert calibration modules and the Sentaurus ML Calibration Accelerator to accelerate the calibration workflow by over 5X. ML enhancements in SCW allow users to create their own modules and workflows, improving TCAD accuracy and reducing calibration time.

The Smart Advantage: How Artificial Intelligence Is Transforming Inspection And Metrology In Semiconductor Manufacturing

Artificial Intelligence (AI) is revolutionizing semiconductor inspection and metrology by enhancing defect detection processes with automation, speed, and adaptability. AI-driven systems leverage Big Data to uncover patterns and anomalies that traditional methods may miss, leading to improved accuracy and efficiency. AI-integrated platforms like Nordson's SQ3000 Multi-Function System can detect microscopic flaws with unparalleled speed and efficiency, surpassing traditional methods. AI's real-time, in-line inspection capabilities enable rapid data processing without compromising production speed, while machine learning models adjust quickly to new production requirements. The advancement of Machine Learning (ML) in inspection systems is transforming defect detection by creating self-teaching AI systems that become smarter and more adaptable with each interaction.

Researchers train living rat neurons to perform real-time AI computations — experiments could pave the way for new brain-machine interfaces

Researchers in Japan trained living rat cortical neurons to generate complex temporal signals using a real-time machine learning framework, integrated with microelectrode arrays and microfluidic devices. The closed-loop system learned to produce periodic and chaotic waveforms without external input, cycling every 333 milliseconds. By confining neuronal cell bodies in specific patterns, the network's dynamics were enhanced, enabling the system to generate various waveforms and approximate a chaotic trajectory. The study suggests that living neuronal networks could serve as novel computational resources, with potential applications in brain-machine interfaces and neuroprosthetic devices.

Microsoft is automatically updating Windows 11 24H2 to 25H2 using machine learning

Microsoft is automatically updating Windows 11 from version 24H2 to 25H2 for Home and Pro users, excluding IT departments and organizations. Users have limited control over the timing of the update but can postpone it temporarily. The forced update is aimed at streamlining future updates and focusing resources on the newer version, as version 24H2 will reach end-of-life in 2026. While Microsoft claims this is a "machine learning-based intelligent rollout," details on this process are not provided, raising concerns about user autonomy and privacy. Microsoft's efforts to improve Windows 11 include plans for enhancements in Windows search, with more improvements expected in the future.

OpenAI is throwing everything into building a fully automated researcher

OpenAI is shifting its focus to building an AI researcher, aiming to create a fully automated system capable of tackling complex problems independently. The company plans to develop an autonomous AI research intern by September, leading to a multi-agent research system by 2028. OpenAI's chief scientist, Jakub Pachocki, believes in the potential of AI models to work autonomously for extended periods, with the goal of applying AI tools to real-world problem-solving. However, concerns about the risks and ethical implications of autonomous AI systems remain, prompting discussions on oversight and control mechanisms.

Exploring the future of Artificial Intelligence — today's models, tomorrow's agents, and the big privacy problem

The article delves into the current state and future of Artificial Intelligence (AI), highlighting the rapid advancements in AI models and their impact on various industries. It discusses the challenges and risks associated with AI, such as hallucination, knowledge uncertainty, and overconfidence in answers. The piece also explores the evolution of AI models, including their reasoning capabilities, multi-modality, and training set sizes, emphasizing the importance of trust in AI outputs. Furthermore, it touches on popular AI models from different vendors like OpenAI, Google, Anthropic, xAI, and Mistral AI, showcasing their unique features and improvements. The article concludes by addressing the privacy concerns surrounding AI, the potential integration of AI into software ecosystems, and the growing influence of AI agents in shaping the future of technology.

DeepSeek tests “sparse attention” to slash AI processing costs

DeepSeek, a Chinese AI company facing export restrictions on advanced AI chips, has developed "DeepSeek Sparse Attention" (DSA) to enhance processing efficiency in its latest language model, DeepSeek-V3.2-Exp. This technique, similar to sparse transformers used by OpenAI and Google Research, aims to reduce computational costs. DeepSeek claims its implementation achieves "fine-grained sparse attention" and has cut API prices by 50%. The company's focus on optimizing performance with limited resources highlights the ongoing efforts to enhance AI models while managing processing costs.

DeepSeek’s new AI model debuts with support for China-native chips and CANN, a replacement for Nvidia's CUDA — Chinese chipmakers Huawei, Cambricon, and Hygon get first-class support

DeepSeek has unveiled its latest AI model, DeepSeek-V3.2-Exp, optimized for Chinese chips and CANN, a CUDA replacement. The model aims to reduce costs for long-context inference with a sparse attention mechanism. Chinese chipmakers like Huawei, Cambricon, and Hygon are actively supporting the model for immediate deployment on their hardware. This move signals China's commitment to AI sovereignty by prioritizing domestic platforms over Nvidia's CUDA ecosystem. The model's compatibility with both Chinese and Nvidia accelerators highlights the country's readiness for a future less reliant on Nvidia hardware.

Will We Know Artificial General Intelligence When We See It?

The article discusses the challenges in identifying Artificial General Intelligence (AGI) and the need for new benchmarks to measure it. With advancements in AI technology, the timeline for achieving AGI has shortened, leading major AI labs to predict its arrival within a few years. Various tests, like the ARC benchmark, are being developed to assess AGI capabilities, focusing on fluid intelligence and the ability to acquire new skills easily. However, defining and testing AGI remains complex due to differing opinions on intelligence, the diversity of human abilities, and the limitations of current benchmarks. Researchers are exploring different benchmarks and tests to evaluate various aspects of AGI, but the ultimate test lies in observing AI's real-world applications and impact.

AI-driven search engine running inside a laundry room aims to rival Google, and you can try it yourself — programmer harnesses old server parts and AI to deliver quality results

Programmer Ryan Pearce has created two search engines, Searcha Page and Seek Ninja, with over 2 billion entries each, aiming to rival Google Search. Using a 32-core AMD EPYC 7532 processor, Pearce's self-hosted search engine sits in his laundry room due to the heat generated. He employs machine learning algorithms to enhance search queries and provide relevant results efficiently. Pearce has written over 150,000 lines of code for the search engine and is considering moving it to a data center-like facility in the future.

Machine Learning Tests Keep Getting Bigger

Nvidia has outperformed competitors in machine learning tests by topping MLPerf's new reasoning benchmark with its Blackwell Ultra GPU in a GB300 rack-scale design. This achievement showcases Nvidia's continued dominance in the field of artificial intelligence and machine learning.

Nvidia claims software and hardware upgrades allow Blackwell Ultra GB300 to dominate MLPerf benchmarks — touts 45% DeepSeek R-1 inference throughput increase over GB200

Nvidia has achieved a 45% increase in inference performance over its previous GB200 platform with the Blackwell Ultra GB300 NVL72 system in MLPerf benchmarks, showcasing hardware and software advancements. The Blackwell Ultra architecture, powering RTX 50-series graphics cards, offers top-tier performance for gaming and AI applications. Nvidia's enhancements include more capable tensor cores, software optimizations, and quantization techniques to improve throughput and maintain accuracy. The company positions the GB300 as economically disruptive for AI factory development, with shipments set to begin soon.

Report: The Road to Artificial General Intelligence: Achieving the Next Era of Intelligence

Industry leaders are exploring the path to achieving Artificial General Intelligence (AGI) in a report by MIT Technology Review and Arm. The report delves into the accelerating timelines for AGI, the need for smarter compute strategies, and the limitations of current AI models in achieving true intelligence. It emphasizes the importance of improving benchmarks, developing new architectures, and adopting innovative approaches to reach AGI. The report targets engineers, researchers, and technology leaders invested in the future of AI.

New AI model turns photos into explorable 3D worlds, with caveats

Tencent has unveiled HunyuanWorld-Voyager, an AI model that transforms photos into 3D-like video sequences, allowing users to navigate virtual scenes. The model generates RGB video and depth data to create consistent 3D reconstructions without traditional methods, though it's not a substitute for video games. While the output isn't true 3D, it simulates camera movement through a 3D space, producing 49-frame video clips that can be linked for longer sequences. Users can define camera paths for exploration, and the system uses a "world cache" to blend image and depth data for realistic video output.

Hardware Technologies And Algorithms for Vector Symbolic Architectures (Purdue Univ., Georgia Tech)

Researchers from Purdue University and Georgia Institute of Technology have published a technical paper on "Cross-Layer Design of Vector-Symbolic Computing," focusing on the convergence of hardware and algorithms in Vector Symbolic Architectures (VSAs). The paper aims to bridge the gap between theoretical software-level explorations and the development of efficient hardware architectures for VSAs. It discusses principles of vector-symbolic computing, hardware technologies for VSAs, and a methodology for cross-layer design. The paper also proposes a hierarchical cognition hardware system as a demonstration of the co-design approach. Open research challenges for future exploration are also highlighted.

Machine Learning In Semiconductor Manufacturing

Machine learning is crucial in AI advancements and can be used in semiconductor manufacturing for predictive maintenance, reducing downtime. However, ensuring data quality and organization is essential for success. Jon Herlocker from Tignis (now part of Cohu) discusses challenges in data gathering, the need for significant computing power, and strategies for maintaining relevant data. This article is part of a series on AI in manufacturing.

Learning More With Less

Smart Glasses Help Train General-Purpose Robots

The AI Agents of Tomorrow Need Data Integrity

Is AI really trying to escape human control and blackmail people?

In June, reports emerged of AI models "blackmailing" engineers and defying shutdown commands, but these incidents were part of contrived testing scenarios. The sensational headlines mask the reality of design flaws rather than AI's intentional malice. These occurrences highlight the need for better understanding of AI systems and human engineering to prevent premature deployment. Just like a malfunctioning lawnmower doesn't intentionally harm, AI models' actions are often due to faulty programming or sensors, not conscious decision-making. The complexity of AI models can lead to misinterpretation of their actions as intentional, when in fact, they are software tools devoid of human-like intentions.

How AI’s Sense of Time Will Differ From Ours

Five ways that AI is learning to improve itself

Mark Zuckerberg aims for smarter-than-human AI at Meta Superintelligence Labs, focusing on self-improving AI systems. AI's ability to enhance itself sets it apart from other technologies, with potential benefits like liberating humans from mundane tasks but also posing risks like rapid advancement in hacking and manipulation. AI is already contributing to its own development in various ways, such as enhancing productivity, optimizing infrastructure, automating training, perfecting agent design, and advancing research. The acceleration of AI progress raises questions about the impact of self-improving AI and the potential for superintelligent models.

A glimpse into OpenAI’s largest ambitions

OpenAI is balancing being a tech giant with a focus on creating "artificial general intelligence" for the benefit of humanity. The company aims to go beyond chatbots to tackle the big questions in AI, such as reasoning like a human and societal implications. Recent successes in coding competitions and math Olympiads show AI's strength in analytical tasks but not in human-like creativity. OpenAI is investing heavily in developing AI models that can reason like humans, aiming to bridge the gap between machine-like reasoning and creative thinking. The company's ambitions include AI that can potentially replace politicians, challenging traditional views on AI capabilities.

Video Friday: Dance With CHILD

Google DeepMind’s new AI can help historians understand ancient Latin inscriptions

Google DeepMind has introduced Aeneas, an AI tool that aids historians in deciphering ancient Latin inscriptions by analyzing partial transcriptions and scanned images to provide possible origins, dates, and missing text fill-ins. Aeneas cross-references text with a database of 150,000 inscriptions to offer parallels for further analysis. The tool aims to assist researchers in generating hypotheses and improving accuracy in determining the origins of inscriptions. While Aeneas has shown promising results, its effectiveness on more obscure samples remains to be seen. The tool is now open-source and freely available for educational and academic use.

DeepMind’s Quest for Self-Improving Table Tennis Agents

A major AI training data set contains millions of examples of personal data

A major AI training data set, DataComp CommonPool, contains millions of personal data examples, including images of passports, credit cards, and birth certificates, according to new research. The study revealed thousands of images with identifiable faces and identity documents within CommonPool, estimating hundreds of millions of such images in the dataset. The data set, released in 2023, consists of 12.8 billion image-text pairs and is used for training generative text-to-image models. Concerns were raised about the presence of personally identifiable information in the data set, highlighting privacy risks and the challenges of filtering such data effectively. Researchers emphasize the need for the machine-learning community to address privacy issues and reconsider the practice of indiscriminate web scraping.

AI Cameras Change Driver Behavior at Intersections

Andrew Ng: Unbiggen AI