AI & Machine Learning

ATLAS-RTC introduces token-level runtime control that detects and corrects LLM drift from structured output contracts during the forward pass.

Guardrails successfully implements and flight-tests Control Barrier Functions on an F-16 fighter jet to enforce safety limits in real-time.

ITQ3_S achieves high-fidelity 3-bit LLM inference by using rotation-domain smoothing to eliminate the catastrophic precision loss caused by outliers.

The Physics-Guided Transformer (PGT) embeds physical priors (like diffusion and causality) directly into the self-attention mechanism via heat-kernel biases.

Iterative Motion Imitation enables bicycle robots to perform unassisted front-flips by learning from initially 'impossible' reference motions.

Proteina-Complexa unifies generative flow-based modeling with structure-based 'hallucination' to set a new SOTA in atomistic protein binder design.

ExFusion enables Transformer models to gain the capacity of Mixture-of-Experts during training while remaining a standard dense model for deployment.

SARL improves reasoning models by rewarding the 'topology' of thoughts rather than just the final answer, enabling effective RL without ground-truth labels.

Dataset Concentration (DsCo) achieves nearly lossless dataset reduction by aligning distributions via diffusion models, cutting storage and training costs by half.

Correlated Diffusion replaces independent noise with structured MCMC dynamics, enabling generative modeling on hyper-efficient probabilistic computers.

This study challenges the common 'best practice' of atomic decomposition for LLM judges, showing that holistic evaluation is often superior at detecting incompleteness.

An autonomous agent reveals that domain-specific molecular architectures are largely unnecessary; standard transformers with better tuning outperform custom designs.

Decoupled language models reduce the compute required for OCR domain adaptation by 95% while matching SOTA transformer accuracy.

This paper clarifies that Diffusion Maps (DMAPs) are not actually a dimensionality reduction tool, but rather a spectral representation that requires specific combinations to form a chart.

The first framework for bit-identical deep learning training that produces MD5-verified identical weights across independent runs.

Drift-AR enables single-step (1-NFE) high-fidelity image generation by reinterpreting AR prediction entropy as a physical drifting field.

Meta-Harness automates the engineering of the 'code' surrounding LLMs, improving RAG and agent performance by optimizing retrieval and context management logic.

ROVED reduces the expensive human feedback required for preference-based RL by up to 90% by leveraging vision-language embeddings and uncertainty filtering.

PhysNet embeds physical tumor growth dynamics directly into the latent feature space of a CNN, rather than just as a constraint on the output.

This paper proves that reward hacking is a structural equilibrium of optimized AI agents, not a bug, and provides a computable 'distortion index' to predict it.

Moves VLM grounding from text-based coordinates to a direct visual token selection mechanism via special pointing tokens.

Introduces Heddle, a trajectory-centric system that resolves the long-tail latency bottleneck of tool calls in agentic Reinforcement Learning.

Bypasses expensive formal verification solvers by designing neural networks that are 'verifiable by design' using the fast trivial Lipschitz bound.

A training-free metacognitive framework that gives LLMs explicit control over expanding, pruning, and repairing reasoning trajectories during inference.

Presents PReD, the first foundation model and 1.3M-sample dataset specifically for electromagnetic signal perception and decision-making.

Replaces traditional fixed-update rules in online learning with a causal Transformer to track switching experts in non-stationary environments.

Replaces the classic Newton-Raphson power-flow solver with a differentiable GPU-accelerated simulation.

Transitions reasoning model optimization from coarse sequence-level advantages to fine-grained token dynamics.

Moves beyond next-token prediction to model reasoning as gradient-based energy minimization over latent trajectories.

Introduces lightweight equilibration to the Muon optimizer, significantly stabilizing and accelerating LLM pretraining.

Discovers that LLM hidden states undergo geometric 'warping' at digit-count boundaries, mimicking human psychological perception.

Enables instruction-following in low-resource languages by simply merging target language base models with English-instructed models.

Enhances Kolmogorov-Arnold Networks (KAN) with fractal interpolation to approximate non-smooth and rough functions.

Exposes a massive robustness gap in Vision-Language-Action (VLA) models, where simple paraphrasing causes up to 50% success drops.

An evolutionary framework for GPU kernel generation that outperforms frontier models like Claude 4.6 and Gemini 3.0.

HISA eliminates the quadratic O(L²) bottleneck in sparse attention indexers, enabling efficient long-context scaling for models like DeepSeek-V3.

Researchers have used LLMs to evolve entirely new Reinforcement Learning update rules from scratch that compete with human-designed baselines like PPO and SAC.

The 'Scaffold Effect' reveals that Vision-Language Models in clinical settings often fabricate reasoning based on prompt framing rather than actual visual data.

Entropic Claim Resolution (ECR) shifts RAG from retrieving 'relevant' documents to retrieving 'discriminative' evidence that minimizes hypothesis uncertainty.

IsoQuant leverages SO(4) isoclinic rotations to achieve a 4.5x-4.7x speedup in low-bit KV-cache quantization over existing methods.

The 'Bidirectional Coherence Paradox' demonstrates that LLM performance and explanation quality can be inversely correlated depending on domain observability.

COvolve creates an automated curriculum for open-ended learning by co-evolving environments and policies as executable code through a zero-sum game.

INSID3 achieves state-of-the-art one-shot image segmentation using only frozen DINOv3 features without any training, fine-tuning, or auxiliary models.

EdgeDiT provides a hardware-aware blueprint for running massive Diffusion Transformers (DiT) on mobile NPUs with a 1.6x reduction in latency.

LAD achieves 3x lower latency than previous driving language models by generating textual reasoning and motion plans at up to 20 Hz.

The TAG glove system provides high-resolution tactile feedback and precise 21-DoF motion capture for under $1000.

Seen2Scene is the first flow matching model trained directly on incomplete real-world 3D scans rather than synthetic complete data.

Hydra unifies ColBERT-style retrieval and autoregressive generation into a single Vision-Language Model using a single LoRA adapter.

StreamingVLA eliminates execution halting in robots by asynchronously parallelizing observation, generation, and execution.

Unrestrained Simplex Denoising treats discrete data generation as a non-Markovian process on the probability simplex.