Juju Radar 论文

Advancing subseasonal forecasting with machine learning

Thu, 15 Oct 2026 09:00:00 +0800

PBC对已有物理或AI天气模型的概率分布进行后处理：Debias++校正地点日期分位偏差，Persistence++结合近期观测与气候规律，再约束并合并有效分布。全球1.5度网格的第三第四周评测和2025实时竞赛提供两类证据，原文方法、图注和代码声明已读。

重要度8.4/10。AIFS平均基线约0.03的RPSS翻倍是相对气候基准的概率技能，并非普遍准确率翻倍。MicroDuet是两种校正预测的组合；依赖历史预测和观测，具体地区和时段仍需验证。

原文：https://www.nature.com/articles/s41586-026-11126-x
代码：https://github.com/mouatadid/pbc
本期资料页：https://www.asteronline.cn/juju-radar/2026-10-15/
AI辅助整理。

Juju Radar 论文

Advancing subseasonal forecasting with machine learning

GreenWAM: A World Foundation Model for Robotics

Towards Automating Scientific Review with Google’s Paper Assistant Tool

Interpretable AI model reveals spatial codes of cell recruitment in tissues

TokenRouter: Efficient Serving System for Token-Level LLM Routing

OneSearch-VL: Unified Multimodal Deep Research Agent for Image and Video

Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?

AgentGarten: Code Worlds for Evolving Agents

Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement

Unified Eukaryotic Genome Annotation: From Gene Prediction to Carbon Annotation Database

Long-WAM: Scaling the Context of World-Action Models

Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation

textGrain: Entropy-Calibrated Watermarking for Language Model Text

AutoGUIWorld: Image Generators as Visual World Models for GUI Agent

ROWBench: Do Video Models Render What the Program Specifies?

Toward provably private learning from federated data

The Strict Threshold for Gaussian Ellipsoid Fitting

Finite-Time Blow-Up of Radial Negative-Energy Solutions for the Mass-Critical Biharmonic Nonlinear Schrödinger Equation

RoboCoach: World Models as Active Coaches for Compositional Robot Skills

Function-preserving watermarking of AI-generated proteins

Scalable decision-making for games of imperfect information

Recursive Harness Distillation across Agents for Robot Manipulation

Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI

InternW0-Δ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data

Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures

Coding Agents for Generalized Task and Motion Planning Problems

Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs

Training Object Permanence in World Models

StudentBench: AI and human tutoring yield equivalent GRE learning gains

Autonomous AI agents discover reverse transcriptases with tandem repeat arrays

MentalHealthBench: An Expert-Informed Benchmark of AI Capabilities in Realistic Mental Health Conversations

RRSI: Regularized Recursive Self-Improvement of Agent Harnesses

RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents

CodeMidas: Scaling Agentic Coding RL Environments from Code Itself

MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations

SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness

JEPA-Anything: Learning Predictive Models across Different Worlds

Accelerating open-source biomolecular models with Claude

Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations

Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control

The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction

Reimagining research papers as interactive and reliable AI agents

Modality-Autoregressive World-Action Models

ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals

Atria Dawn: The Dawn of Agentic Superintelligence

Learning to Solve Hard Problems in RL for LLMs by Never Giving Up

ModaLens: Measuring Image Sensitivity in Report-Conditioned Medical VLMs

Breaking the Vision-Action Shortcut: Latent Interface Training for Generalizable Robotics Foundation Models

SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking

SNAP3D: Physically Grounded 3D Parts for Assembly from a Single Image

Mi-Ripple: Restoring Images Degraded by Iterative AI Editing

Beyond Solver Verdicts: Generative Reward Models for Autoformalization

X-AuT: Progressive Audio-Encoder Compression for Speech LLMs with Cross-Scale Distillation

Xiaomi-CocktailASR-1 Technical Report

An Open Recipe for IMO Gold: Training Nemotron for Olympiad Mathematics

NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

SenseNova-U1.5: Towards Native Unified Visual Intelligence

Memory as Plans: World-Action Modeling with Memory-Grounded Planning

Negative Self-Distillation: Learning to Reason by Avoiding Flaws

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

Show-Harness: Just a VLM Agent Can Play Robots

Programmable World Model

Detecting and countering misuse of AI: September 2026

AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing

OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining

Predicting genome-wide functional constraints with GPN-Star

Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation

Finite time blowup for Navier–Stokes

Unlocking Lossless Speedups in LLMs via Discrete Diffusion

τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction

RISE: Recursive Improvement via Self-Extrapolating Policy Distillation

WorldSculpt: Generating Compositional Worlds from Grounded Videos

Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment

Principia: Relational Physics Tests for Video Models

Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning

Editable Visual Design

LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes