Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers introduced Geometry-Aware Tabular Diffusion (GATD), a method that improves tabular data synthesis by feeding pairwise angles and lengths from column value differences into diffusion denoisers as inputs and au…
Researchers found that a linear probe applied to mid-layer hidden states of quantized large language models can detect hallucinations with up to 1.000 AUROC, significantly outperforming sampling-based methods that max ou…
Researchers have introduced GAMBLe, a new analytical framework that decomposes AI-Driven Research Systems (ADRS) into four key parameters and one compositional object to better understand their behavior. Testing across 7…
Researchers have developed a collaborative inference method for occlusion-robust object detection on edge devices with less than 1 MB SRAM, combining an MCUNet backbone, YOLOv2 detection head, and TensorFlow Lite quantiz…
Researchers found that visual graph scaffolds significantly improve large language models' multi-hop reasoning compared to flattened text representations of graph structures. In experiments on question-answering tasks, g…
Researchers have developed AURA-Mem, a memory system for robot policies that maintains a constant 4,224-byte inference state regardless of episode length, compared to a KV-cache that grows 6,061 times larger over 100,000…
Researchers have developed ChatHealthAI, a multimodal reasoning framework that aligns structured electronic health record (EHR) representations from a pretrained foundation model with the semantic space of a frozen large…
Researchers introduced Reasoning Primitive Induction, a single-pass method that mines successful ReAct agent traces to extract recurrent reasoning moves into a compact library of typed pseudo-tools. The induced libraries…
A new study on Large Reasoning Models (LRMs) has found that continuing to reason after reaching a correct answer can cause the model to deviate from that answer, a phenomenon termed "harmful overthinking." Researchers in…
Researchers introduced Mixture of Pathway Experts (MoPE), a knowledge-distillation framework that uses RNA-derived pathway information during training to improve computational pathology models that only analyze whole-sli…
Researchers have developed Adaptive Latent Agentic Reasoning (ALAR), a dual-mode framework that reduces unnecessary textual reasoning in LLM agents by using compact latent reasoning for routine tasks and reserving explic…
NVIDIA researchers introduced Cosmos 3, a family of omnimodal world models that jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. The …
Researchers introduced ToolGate, a lightweight external controller that predicts whether to execute or skip tool calls proposed by vision-language agents, addressing the pre-call control problem where many tool execution…
Researchers introduced the Ghost Annotator framework, which combines conformal prediction with collaborative filtering to model how large language models (LLMs) behave compared to human annotators in content moderation t…
Researchers introduced GeoDrive-Bench, a benchmark of 5,053 human-validated multiple-choice questions across six countries to test vision-language models on region-specific traffic rules and driving behavior. Testing nin…
Researchers introduced AVTrack, a new human-centric audio-visual instance segmentation dataset designed for dynamic real-world scenarios. The dataset features challenging conditions such as camera motion, visual occlusio…
Researchers have identified a critical flaw in using standard survival prediction models for organ allocation, showing that even highly accurate predictors optimized for metrics like the C-index can perform no better tha…
Researchers have developed a method to predict how much accuracy improves when using best-of-N inference scaling in language models, without running the full procedure. By analyzing statistics from a single labeled valid…
Researchers have developed a Marchenko-Pastur (MP) random-matrix approach for pruning deep neural networks that requires minimal post-pruning fine-tuning. The method provides deterministic data-path certificates for accu…
A systematic human audit of the NL-to-FOL benchmarks FOLIO and MALLS found that approximately 39% and 36% of entries, respectively, contain incorrect FOL formalizations, with additional errors in ambiguous natural langua…