Official platform of Tsinghua Big Data Research Center, sharing the team's latest research, teaching updates, and big data news.
Researchers used Claude Science AI to combine multi-wavelength astronomical data and fill gaps in NASA's GALEX ultraviolet survey, producing the first complete all-sky UV map with 10% prediction accuracy and revealing structures like stellar nurseries and supernova remnants.
This 72-page survey explores how optimal transport provides geometric tools for comparing distributions in reinforcement learning, covering imitation learning, policy optimization, model-based RL, offline RL, robustness, and transfer, while analyzing design choices in cost functions, temporal handling, and computational trade-offs.
This survey paper introduces a four-axis taxonomy for Retrieval-Augmented Generation (RAG) systems—efficiency, defense, interactivity, and reasoning—moving beyond simple vector database integration to analyze design trade-offs in evidence retrieval, security, user alignment, and multi-step reasoning.
This study evaluates nine open-source LLMs on cross-country value simulation using World Values Survey data across 59 countries, introducing representational equality as a metric to assess whether simulation accuracy is evenly distributed across nations, and tests four intervention strategies for improving both accuracy and equality.
This review examines how evolutionary computation supports trustworthy AI across three research lines: evolutionary attacks to find failures, evolutionary defenses to search protective measures, and trustworthy self-evolution to govern which changes persist long-term, highlighting the shift from static robustness to continuous governance of memory, skills, and safeguards.
Google's TimesFM-3 introduces native multivariate time series forecasting with alternating causal temporal and full variate attention, non-autoregressive decoding via contiguous patch masking, and support for historical and future-known covariates, achieving top rankings on Gift-Eval, FEV-Bench, and TIME benchmarks without sacrificing univariate performance.
The PsychAD consortium analyzed 6.3 million single-nucleus transcriptomes from 1,494 donors spanning infancy to 108 years, creating the largest prefrontal cortex atlas to date and revealing shared and disease-specific molecular changes across eight brain disorders using AI-driven graph neural networks.
NASA and IBM release an open-source lunar foundation model trained on 17 years of Lunar Reconnaissance Orbiter data across 11 modalities and two resolutions, enabling detection of polar ice, young volcanic structures, and craters with performance matching or exceeding baselines, integrated into the TerraTorch toolkit for reproducible research.
UT Austin PhD student Cao Jin questions whether computer science academia is obsolete as foundation models absorb specialized research, arguing that incremental SOTA-chasing papers are becoming obsolete and researchers must either join industry to build models or pivot to fundamental science like mechanistic interpretability.
The article explains how memory poisoning attacks persist across sessions in AI agents, details the memoryfield format for auditable storage, proposes a secure write path with trust levels and source tracking, and provides detection queries for identifying poisoned memories.
