Official platform of Tsinghua Big Data Research Center, sharing the team's latest research, teaching updates, and big data news.
The article details how the open‑source Flash‑MoE engine enables the 400B‑parameter Qwen3.5‑397B‑A17B mixture‑of‑experts model to run on an iPhone 17 Pro, achieving about 0.6 tokens per second through a custom Metal pipeline, GCD‑driven SSD streaming, and aggressive caching strategies.
This article reviews the five‑year evolution of reinforcement‑learning techniques for large language models, comparing PPO, DPO, GRPO and emerging multi‑agent approaches, analyzing their reward signals, practical trade‑offs, and the open‑source frameworks that support them.
The UniCM model introduced by Tsinghua University's Li Yong team unifies learning of multiple ocean‑atmosphere climate modes with a dual‑branch Transformer, achieving record‑long ENSO forecasts and revealing hidden inter‑modal couplings that turn AI from a fast weather predictor into a climate discovery tool.
AutoScientists is a self‑organizing multi‑agent framework that automates the full scientific loop—from hypothesis generation to paper writing—demonstrating superior performance on BioML‑Bench (74.4% average rank, +8.33% over baselines) and achieving notable gains in protein‑engineering tasks such as ACE2‑Spike binding.
The article details how an AI‑augmented workflow completed a 46‑page research paper in six days using 108 agent calls and 648 k tokens, introduces an L1‑L5 autonomy taxonomy, compares four architectural patterns across 17 systems, and highlights six open challenges and key bottlenecks such as continual knowledge accumulation and reliable self‑assessment.
The article examines how recent advances in large language models, AutoML, and evolutionary algorithms are pushing AI toward recursive self‑improvement, outlines current capabilities and limitations, and discusses the technical, economic, and safety challenges that still prevent a fully autonomous intelligence explosion.
ESMFold2, an open‑source protein‑language‑model predictor, now offers predictions for 1 billion structures and a 6.8 billion‑sequence Atlas, delivering faster, more accurate antibody and protein‑interaction designs than AlphaFold3, while demonstrating experimental success and revealing novel CRISPR‑like proteins, though atypical cases still need careful validation.
This article reviews two ACL 2026 papers that explain why steering works for large language models, introduce a three‑stage behavior model and activation‑manifold hypothesis, propose the SPLIT method, present the SteerEval evaluation framework, and describe the EasyEdit2 open‑source toolkit.
A reinforcement‑learning framework called SOFisher, developed by teams from Fudan and Beijing Institute of Technology, enables intelligent, adaptive selection of field‑of‑view positions in costly spatial‑omics experiments, dramatically improving target detection efficiency and revealing disease‑relevant cellular niches with far fewer measurements.
The article introduces the ICML 2026 Theory of Agent (ToA), analyzes four common failure modes of modern agents, explains the internal‑vs‑external tool trade‑off through a knowledge‑boundary framework, and outlines how effort‑conservation and the β parameter guide self‑evolving agent design and future research.
