Close Menu
Versa AI hub
  • AI Ethics
  • AI Legislation
  • Business
  • Cybersecurity
  • Media and Entertainment
  • Content Creation
  • Art Generation
  • Research
  • Tools

Subscribe to Updates

Subscribe to our newsletter and stay updated with the latest news and exclusive offers.

What's Hot

Benchmarking large-scale language models for healthcare

June 8, 2025

Oracle plans to trade $400 billion Nvidia chips for AI facilities in Texas

June 8, 2025

Research papers provide a roadmap for AI advancements in Nigeria

June 7, 2025
Facebook X (Twitter) Instagram
Versa AI hubVersa AI hub
Sunday, June 8
Facebook X (Twitter) Instagram
Login
  • AI Ethics
  • AI Legislation
  • Business
  • Cybersecurity
  • Media and Entertainment
  • Content Creation
  • Art Generation
  • Research
  • Tools
Versa AI hub
Home»Research»Research focus: Week of December 16, 2024
Research

Research focus: Week of December 16, 2024

By December 18, 2024Updated:February 13, 2025No Comments6 Mins Read
Share Facebook Twitter Pinterest LinkedIn Tumblr Reddit Telegram Email
Share
Facebook Twitter LinkedIn Pinterest Email

Welcome to Research Focus. This is a series of blog posts highlighting notable publications, events, code/datasets, new hires, and other milestones from across the Microsoft research community.

new research

NeoMem: Hardware/Software Co-Design for CXL Native Memory Tiering

The Compute Express Link (CXL) open standard interconnect allows you to integrate different types of memory into your servers through byte-addressable SerDes links. To take full advantage of CXL-based heterogeneous memory systems (systems that combine different types of memory with different access speeds), efficient memory tiering, that is, a system that combines different types of memory with different access speeds, requires efficient memory tiering, or You need to implement a strategy to manage data placement. Efficiently managing these memory systems is critical, but this has been difficult due to the lack of accurate and efficient tools to understand how memory is accessed. .

In a recent paper, “NeoMem: Hardware/Software Co-Design for CXL-Native Memory Tiering,” Microsoft researchers propose a new solution featuring hardware/software co-design to address this problem. Masu. NeoMem offloads memory profiling functionality to the CXL device-side controller and integrates a dedicated hardware unit called NeoProf. NeoProf monitors memory access and provides important page hotness statistics and other system state information to the operating system (OS). On the OS kernel side, the researchers designed an improved memory tiering strategy to enable accurate and timely hot page promotion based on NeoProf statistics. NeoMem was implemented on a real FPGA-based CXL memory platform and Linux kernel v6.3 and demonstrated geo-average speedup of 32% to 67% compared to several existing memory tiering solutions.

new research

Chimera: Accurate retrosynthesis predictions with ensemble models with diverse inductive biases

Planning and execution of chemical synthesis is a key challenge in the discovery of functional small molecules and limits the potential of generative AI in molecular reverse design. Early machine learning-based retrosynthesis models have shown the ability to predict reasonable routes, but with low accuracy for infrequent but important reactions.

In a recent paper, “Chimera: Accurate retrosynthesis predictions with ensemble models with diverse induced biases,” Microsoft and external researchers use a new framework for building highly accurate reaction models to This limitation is addressed. Chimera incorporates two newly developed models that deliver cutting-edge performance in their respective categories. Evaluations by PhD-level organic chemists indicate that chimera predictions are preferred due to their higher quality compared to the baseline model.

Researchers further validated Chimera’s robustness by applying Chimera’s largest model to an internal dataset at a major pharmaceutical company, demonstrating Chimera’s ability to generalize effectively under changing distributions. . This new framework shows the potential to significantly accelerate the development of more accurate and versatile reaction prediction models.

Spotlight: Event Series

microsoft research forum

Join us for an ongoing exchange of ideas about general AI-era research. Watch the first four episodes on demand.

opens in new tab

new research

GA4GH Task Execution API: Enables easy multi-cloud task execution

In bioinformatics and computational biology, data analysis often involves chaining together command-line programs developed by specialized teams at different institutions. These tools vary widely in age, software stacks, dependencies, and lack a common programming interface, making integration, workflow management, and reproducibility difficult.

recent articles (Opens in new tab) Highlights the development, adoption, and implementation of the Global Alliance for Genomics and Health (GA4GH) Task Execution Services (TES) API, created in collaboration with researchers at Microsoft and other institutions. The TES API provides a unified schema and interface for task submission and management, seamlessly bridging the gap between on-premises high-performance and high-throughput computing systems, cloud platforms, and hybrid infrastructures. Its flexibility and scalability have already made it a valuable asset for applications ranging from federated data analysis to load balancing across multicloud systems.

TES has been adopted by numerous service providers and integrated into several workflow engines, allowing researchers to perform complex computational tasks through a single abstracted interface. This eliminates compatibility roadblocks, shortens research timelines, reduces costs, and enables “compute to data” solutions essential to tackling the challenges of distributed data analysis.

new research

RedCode: Dangerous Code Execution and Generation Benchmark for Code Agents

The increasing use of code agents in AI-assisted coding and software development raises safety and security concerns, such as the generation and execution of malicious code, which is a major barrier to the practical adoption of these agents. Masu.

In a recent paper presented at NeurIPS 2024, “RedCode: Risky Code Execution and Generation Benchmark for Code Agents,” Microsoft and external researchers propose a comprehensive and practical assessment of the safety of code agents. . RedCode is an evaluation platform with benchmarks based on four key principles: real interaction with the system, comprehensive assessment of unsafe code generation and execution, diverse input formats, and high quality safety scenarios and tests. is.

In this study, we evaluated three agents based on different large-scale language models (LLMs) and provided insights into code agent vulnerabilities. For example, our results show that agents are more likely to refuse to perform unsafe operations on the operating system. Unsafe operations written in natural text have a lower rejection rate than operations written in code form. Additional evaluation reveals that more capable base models and agents with stronger overall coding capabilities, such as GPT-4, tend to produce more sophisticated and harmful software. Ta.

These findings highlight the need for rigorous safety evaluation of diverse code agents. The underlying dataset and associated code are publicly available at https://github.com/AI-secure/RedCode. (Opens in new tab).

new research

Towards industrial infrastructure models: Integrating large-scale language models and industrial data intelligence

Large-scale language models (LLMs) are great for language-focused tasks like news writing, document summarization, customer service, and virtual assistant support, but they’re great for learning tabular or structured industry data. and may face challenges when it comes to reasoning. Time series data. To address these issues, Microsoft researchers are proposing a new approach to building industrial infrastructure models (IFMs). As outlined in a recent blog post, they successfully demonstrated the feasibility of cross-domain universal in-context learning on tabular data and the great potential it can achieve.

Researchers designed generative tabular learning (Opens in new tab) (GTL) is a new framework that integrates zero-shot and few-shot learning capabilities from multiple industries into LLM. This approach allows the model to adapt and generalize more effectively to new fields, new data, and new tasks, making it flexible for diverse data science tasks. This technology paradigm is open sourced (Opens in new tab)Encourage wider use.

Microsoft Research News

A new frontier awaits – computing with light

December 12, 2024

Inside newer types of computers, many small LEDs glow green. Those lights have a job to do. They are doing the math. For now, this mathematics tells computers how to identify handwritten images of numbers. This computer is part of Microsoft’s research program.

See more news and awards

opens in new tab

author avatar
See Full Bio
Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
Previous Article5 AI PCs to upgrade your content creation
Next Article Cineverse launches software to help content owners monetize AI

Related Posts

Research

Research papers provide a roadmap for AI advancements in Nigeria

June 7, 2025
Research

JMU Education Professor was awarded for AI Research

June 3, 2025
Research

Intelligent Automation, Nvidia and Enterprise AI

June 2, 2025
Add A Comment
Leave A Reply Cancel Reply

Top Posts

Deepseek’s latest AI model is a “big step back” for free speech

May 31, 20255 Views

Doudna Supercomputer to Strengthen AI and Genomics Research

May 30, 20255 Views

From California to Kentucky: Tracking the rise of state AI laws in 2025 | White & Case LLP

May 29, 20255 Views
Stay In Touch
  • YouTube
  • TikTok
  • Twitter
  • Instagram
  • Threads
Latest Reviews

Subscribe to Updates

Subscribe to our newsletter and stay updated with the latest news and exclusive offers.

Most Popular

Deepseek’s latest AI model is a “big step back” for free speech

May 31, 20255 Views

Doudna Supercomputer to Strengthen AI and Genomics Research

May 30, 20255 Views

From California to Kentucky: Tracking the rise of state AI laws in 2025 | White & Case LLP

May 29, 20255 Views
Don't Miss

Benchmarking large-scale language models for healthcare

June 8, 2025

Oracle plans to trade $400 billion Nvidia chips for AI facilities in Texas

June 8, 2025

Research papers provide a roadmap for AI advancements in Nigeria

June 7, 2025
Service Area
X (Twitter) Instagram YouTube TikTok Threads RSS
  • About Us
  • Contact Us
  • Privacy Policy
  • Terms and Conditions
  • Disclaimer
© 2025 Versa AI Hub. All Rights Reserved.

Type above and press Enter to search. Press Esc to cancel.

Sign In or Register

Welcome Back!

Login to your account below.

Lost password?