Research
Exploring the latest AI research coming out of Asia's universities and labs.
New AI System Cuts Energy Use by 100x While Boosting Accuracy Through Symbolic Reasoning
A hybrid approach combining neural networks with human-like symbolic reasoning achieves dramatic energy savings and could reshape data center economics.
Mimic Robotics' Video-Action Model Achieves 10x Sample Efficiency in Robot Learning
A new video-action model pairs pretrained internet-scale video with a flow-matching action decoder, enabling robots to learn manipulation tasks with dramatically less real-world data.
Neuromorphic Computers Can Now Solve Physics Simulations Once Reserved for Supercomputers
Brain-inspired processors demonstrate massive energy improvements over GPUs, and a new EU project aims to push the technology further using microscopic LEDs.
KIT Researchers Build AI System That Predicts Scientific Breakthroughs 2-3 Years Ahead
Published in Nature Machine Intelligence, the system maps scientific literature into knowledge networks and identifies which concept combinations will gain significance.
MIT's CompreSSM Uses Control Theory to Compress AI Models During Training, Achieving 4x Speedups
A new technique from MIT CSAIL applies Hankel singular values to identify and remove unnecessary model components while training is still underway.
Z.AI's GLM-5.1: A 754B Open-Weight Model That Tops SWE-Bench Pro and Runs Autonomously for 8 Hours
The Chinese lab's latest release achieves state-of-the-art 58.4% on SWE-Bench Pro under an MIT license, outperforming GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro on agentic coding.
MIT Builds AI Model That Detects Six Types of Atomic Defects in Materials Without Cutting Them Open
The machine learning system, trained on 2,000 semiconductor materials covering 56 elements, uses neutron-scattering data to noninvasively identify defects that could improve heat transfer and energy conversion.
USGS River DroughtCast Uses Century of Data to Predict Water Shortages With Machine Learning
The new forecasting tool covers 3,000+ streamgage locations and bridges the critical gap between short-range weather forecasts and seasonal water supply outlooks.
Hybrid AI Approach Slashes Energy Use 100x While Boosting Accuracy
Researchers combine neural networks with symbolic reasoning to create a dramatically more efficient framework that mirrors human problem-solving.
Carnegie Mellon Launches AI-Astronomy Fellowship to Accelerate Cosmic Discovery
The Simons Foundation-backed Keystone program pairs visiting astrophysicists with AI mentors to tackle frontier problems at the intersection of machine learning and the cosmos.
Neuromorphic Computers Crack Physics Simulations Once Reserved for Supercomputers
Sandia National Labs demonstrates that brain-inspired chips can solve partial differential equations with near-ideal scaling and far less energy.
Google's TurboQuant Algorithm Tackles AI's Memory Wall at ICLR 2026
A novel two-step compression algorithm using PolarQuant vector rotation and quantized Johnson-Lindenstrauss projection dramatically reduces KV cache overhead, potentially shifting AI development toward efficiency-first paradigms.