AI 论文与研究

Research ideation is iterative: researchers refine their ideas step by step based on feedback. Yet existing evaluations obscure this search process, making it difficult to understand how LLMs explore the space of research ideas and what li…

Neural networks are increasingly studied not merely as predictive systems, but as complex systems whose internal computations are themselves objects of scientific inquiry. Recent progress in mechanistic interpretability (MI) has made it po…

In this paper, we present 1020 Group's submission to the WMT26 shared task on automated quality evaluation. Our system, Pterodactyl, was engineered specifically for subtasks 2 and 3, that is, segment-level quality score prediction and the …

Text-to-SQL systems over large pooled catalogs must first retrieve the tables a question needs. We compare against the published table recall of RASL, which uses Claude 3.5 models to extract keywords and select tables, on pooled BIRD, Spid…

Visual Question Answering (VQA) imposes substantial computational overhead on vision-language models (VLMs), motivating model-agnostic input reduction methods that can be readily applied across diverse VLMs, including in black-box settings…

We study Bayesian inference of persistence from early observations of a heterogeneous susceptible--infected--susceptible (SIS) reaction--diffusion model with standard incidence. We show that the leading prevalence dynamics depend on the un…

Policy optimization is central to language model post-training, yet principled guidance remains limited. We revisit its theoretical foundations to identify algorithms that learn near-optimal policies using total samples and gradient steps …

Cross-validation estimates predictive performance by repeatedly splitting the data into training and validation sets, fitting the model on the first and evaluating it on the second. Every observation influences the estimate in two roles, t…

很高兴与 @GoogleDeepMind 合作，将 MKA Benchmark 引入 Artificial Analysis

Artificial Analysis 与 Google DeepMind 合作上线 Multimodal Knowledge Agent（MKA）评测，用 200 个覆盖 10 个行业的私有真实视觉任务考察智能体处理建筑图纸、工程示意图和临床扫描并产出幻灯片、表格、PDF 等交付物的能力。

UC Berkeley 等机构的 Route Anything 研究系统评估模型与策略路由，发现在 25 个基准上，路由平均带来 8.82 倍的单位成本效率提升，质量损失不到 0.5 分。

机器人学习取得了惊人的进展！来看看 #CoRL2026 上的 BEHAVIOR 挑战赛 👇

Philipp Schmid（Google DeepMind 开发者体验）宣布 Gemini 4 Carbon 现已推出，并给出官方博客链接了解更多：https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/。原文仅包含发布链接，未提供模型细节。

Wiz Research 介绍其 Cyber Model Arena v3.9 的基准设计方法，该基准覆盖 26 个模型、59 个智能体和 327 道挑战，分代码漏洞、API 安全、Web 安全和云安全四类。

Robbyant 开源 Apache 2.0 许可的 LingBot-Touch，将指尖传感器读数转为机器人控制软件可用的触觉信号。三个同尺寸编码器各 5.46M 参数，在 4000 小时数据上 JEPA 版本触觉→视频检索 top-5 达 49.3%，MAE 低于 9%，VAE 仅 3.1%。

Google DeepMind 发布 AlphaProof Code 多智能体形式化验证框架，并首次对广泛部署协议 Signal 的生产机器码完成形式化安全验证。

谷歌 DeepMind 团队利用具备自主执行能力的 AI 智能体，对 Signal 加密协议完成形式化验证，用数学方法证明其满足安全性承诺，成果被称为具有里程碑意义的进展。

OpenAI 发布研究，分析预训练、架构、强化学习和模型生成数据如何影响思维链（CoT）可监控性。

小米联合北京大学在 2026 CCF 中国软件大会上发布《物联网设备模型（Thing Model）技术白皮书》，基于超 11.6 亿台 AIoT 设备、200+ 品类和 10000+ 家合作伙伴的实践。

OpenAI 与 Upwork 发布研究报告，分析 AI 对自由职业市场的影响。研究覆盖 63 个工作类别，其中 43 个类别中 ChatGPT 使用较活跃的项目在 Upwork 上的份额下降；AI 相关职位的任务组合到 2025 年底多样性提升 45%，约从 15 项等常见任务增至 21 项；AI 相关项目的固定预算更大、工时更多， hourly 费率相近。

Nvidia 等机构论文发现，提示中的无关文本主要通过削弱模型对问题的利用来降低正确答案的支持度，而非直接压制正确答案，机制可定位到少量 attention heads，即 intrusion heads。

Diabetic retinopathy (DR) lesions vary widely in shape, and microaneurysms (MAs) span only a few pixels. We hypothesize that modeling the spatial organization of lesion regions, with global context and shape-adaptive scanning, improves the…

We propose a novel conditional entropy function as a basis for quantifying predictive uncertainty in Evidential Deep Learning (EDL) for glaucoma detection from retinal fundus images. Using the EyePACS-AIROGS-light-V2 and RIM-ONE-DL dataset…

Nepali speakers regularly mix English words into conversation, but multilingual speech recognisers often struggle with language switches and the writing script used for English words. We study how to adapt SraVaani, a 449M-parameter hybrid…

Remote sensing imagery acquired by Unmanned Aerial Vehicles (UAVs) supports wide range of applications; however, deploying machine learning (ML) models on resource-constrained aerial platforms introduces challenges beyond predictive accura…

Small language models (SLMs) running on mobile devices can keep user data private and work without connectivity. But they perform well only when fine-tuned for the target task and sized for the target device, a process that today requires …

SDR-to-HDR reconstruction is an ill-posed problem in which clipped highlights, crushed shadows, and quantization artifacts must be recovered while preserving the original content. Latent diffusion models provide powerful generative priors …

Compositional generalization is an important form of out-of-distribution generalization, defined as the ability to train on some combinations of modules (e.g., functions such as sort, filter) and then generalize to unseen combinations. Des…

Open-Vocabulary Scene Graph Generation (OV-SGG) requires models to recognize visual relationships beyond the training vocabulary, yet existing methods often rely on dataset-specific object--predicate co-occurrence patterns. Under long-tail…

Simulation optimization typically returns a single estimated optimal point. When decisions also depend on criteria evaluated independently of simulation, such as resource cost, the decision maker instead needs the set of designs whose expe…

Existing Chinese educational corpora often suffer from fragmented coverage and a lack of systematic organization. In this paper, we propose WoW (\textbf{Wo}rld Kno\textbf{w}ledge), a large-scale dataset designed for the full-cycle training…

Reliable inference for rankings of large language models based on human pairwise comparisons must preserve the privacy of individual preference records. To meet this need, we develop methods for score and rank inference under the Bradley-T…

KL-regularized reinforcement learning, the objective behind maximum-entropy RL and RL from human feedback, is variational inference in the graphical model of "control as inference": the regularized optimal policy is a posterior over trajec…

Integrating human feedback into language models has gained increasing attention as a means to align model outputs with human preferences, with Reinforcement Learning from Human Feedback (RLHF) serving as a prominent framework for preferenc…

As AI agents become increasingly capable, users must still recognize when and how to ask for help, leaving many opportunities for useful support unrealized. Proactive agents hold the promise to close this gap, but effective proactivity req…

As AI-based products permeate society, specifically systems using generative models, a belief system is emerging about what these products are and what they can and cannot do. One common belief concerns the perception of the form of langua…

We introduce SafeRoute-MAS, a coverage-aware, risk-calibrated procedure that keeps all specialist messages or removes one using only pre-integration information. On a sealed cache of 5,328 committees, its source-compatible operating point …

Analog integrated circuit (IC) design is a multi-turn, simulator-grounded engineering task that demands iterative reasoning over coupled performance tradeoffs, yet proprietary process design kits (PDKs) make hosted large language model wor…

Active learning aims to reduce annotation cost by selecting the most informative examples for labeling. In the low-budget regime, where uncertainty-based methods often fail, coverage-based active learning methods such as ProbCover have bee…

Parallel drafters accelerate LLM decoding by proposing a whole block of draft tokens in a single forward pass, making them attractive for LLM serving. Recent systems optimize goodput (i.e., accepted tokens per second) because rejected draf…