AWS introduces Parallel-EAGLE (P-EAGLE) to enhance language model inference speed by predicting all speculative draft tokens simultaneously in a single forward pass. P-EAGLE eliminates the sequential drafting phase, delivering up to a 1.69x throughput speedup over traditional frameworks like EAGLE-3, now supported by Amazon SageMaker JumpStart.
Amazon SageMaker AI introduces container image caching to speed up latency by up to 2x during scale-out events, addressing the container image download bottleneck for generative AI models. This advancement improves auto scaling responsiveness, removing the need to download container images when launching new instances, benefiting endpoint scale-out for various AI workloads.
Coherent expands AI manufacturing in Texas with $50 million CHIPS Act grant, boosting US semiconductor production. NVIDIA and Coherent CEOs lead groundbreaking for world's first 6-inch indium phosphide fab, crucial for AI infrastructure.
LangChain Deep Agents addresses the challenge of depth versus context in AI-powered research workflows by delegating deep work to isolated subagents. Amazon Bedrock AgentCore provides the infrastructure needed, allowing developers to build competitive research agents with isolated execution environments for multi-step AI workflows.
Refactored C# kernel ridge regression approximates support vector regression. Technique combines advantages of both methods for efficient large dataset handling.
Databricks releases Omnigent, an open-source 'meta-harness' for AI agents under Apache 2.0 license, enabling seamless collaboration and control. Omnigent standardizes interfaces, allowing engineers to easily swap and coordinate multiple agents, enhancing composition and sharing capabilities.
C# "dynamic" keyword simplifies adding secondary evaluation metrics to regression models, enhancing flexibility and efficiency. Demo showcases diverse evaluation methods like RMSE, R2, and Baseline Accuracy for improved model assessment.
Zyphra introduces Zamba2-VL, a family of open vision-language models with a unique hybrid state-space design for competitive accuracy at lower latency. The Zamba2 backbone combines Mamba2 state-space layers and shared transformer blocks, outperforming other models in benchmarks like PixMoCount and Document understanding.
Rocket Close, a Detroit-based company within Rocket Companies, developed Supercharger, an AI solution in collaboration with AWS to optimize title operations workflows and improve efficiency in the lending and homebuying process. Supercharger centralizes knowledge, automates research-heavy tasks, and enhances both operational efficiency and client experience, powered by Strands Agents and Amazon...
Amazon Bedrock Data Automation (BDA) simplifies structured data extraction from varied documents with customizable blueprints. Blueprint instruction optimization enhances accuracy without separate model fine-tuning, revolutionizing document field extraction.
The Hertz Foundation awarded fellowships to MIT students Annika Marschner, Alvin Q. Meng, Zachary S. Siegel, and Matthew Wanta, providing 5 years of financial support for groundbreaking research. Recipients gain autonomy and access to a network of over 1,300 fellows, leading to collaborative breakthroughs in science and technology fields.
Jinhua Zhao appointed head of MIT's Department of Urban Studies and Planning, known for shaping global mobility systems and bridging research with policy. Zhao's work with leading transportation agencies worldwide and founding of MIT Mobility Initiative highlight his impact on shaping future mobility solutions.
AdaBoost. R2 regression predicts single numeric values by improving tree regressors sequentially. The demo program achieves 82.50% accuracy on training data and 52.50% on test data.
L. L. Thurstone's 1927 paper on random utility models laid the foundation for understanding human preferences. Recent research by MIT experts reveals new insights and potential improvements to these models.
AI agents require evaluation beyond output-level testing to catch failures in tool usage and data fidelity. Agent-EvalKit integrates with AI coding assistants to provide infrastructure for full execution path evaluation, generating targeted test cases and improvement recommendations.