research

From Correlation to Cause: A Five-Stage Methodology for Feature Analysis in Transformer Language Models

arXiv:2605.22462v1 Announce Type: new Abstract: We propose a five-stage methodology for causal feature analysis in transformer language models (probe design, feature extraction, causal validation, rob

Research

From Reasoning Chains to Verifiable Subproblems: Curriculum Reinforcement Learning Enables Credit Assignment for LLM Reasoning

DGX agent

arXiv:2605.22074v1 Announce Type: cross Abstract: Reinforcement learning from verifiable rewards (RLVR) has shown strong promise for LLM reasoning, but outcome-based RLVR remains inefficient on hard p

researcharxiv-cs-cl

22 May 2026

From TF-IDF to Transformers: A Comparative and Ensemble Approach to Sentiment Classification

arXiv:2605.22003v1 Announce Type: new Abstract: Sentiment analysis, also referred to as opinion mining, primarily tries to extract opinion from any text-based data. In the context of movie reviews and

GazePrior: Zero-Shot AR/VR Eye Tracking via Learned 3D Gaze Reconstruction

arXiv:2605.22359v1 Announce Type: new Abstract: Eye tracking (ET) is a foundational technology for advanced AR/VR applications. However, training ET models for every new ET device is challenging: real

Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift

arXiv:2605.21849v1 Announce Type: cross Abstract: Mechanistic interpretability aims to explain a model's behavior by identifying causally responsible internal structures. Dictionary-based explainers s

Google I/O showed how the path for AI-driven science is shifting

During Tuesday’s Google I/O keynote, Demis Hassabis, the CEO of Google DeepMind, proclaimed that we are currently “standing in the foothills of the singularity.” It was a striking statement—the singul

researchmit-tech-review

Guided Trajectory Optimization with Sparse Scaling for Test-Time Diffusion

arXiv:2605.21907v1 Announce Type: new Abstract: The efficient Test-Time Scaling (TTS) paradigm offers a promising perspective for enhancing the generation performance of diffusion models. However, cur

High Quality Embeddings for Horn Logic Reasoning

arXiv:2605.20467v1 Announce Type: new Abstract: Neural networks can be trained to rank the choices made by logical reasoners, resulting in more efficient searches for answers. A key step in this proce

Higher Order Reasoning for Collaborative Communicationless Mobile Robot Operations

arXiv:2605.21901v1 Announce Type: new Abstract: In communicationless environments, multi-robot systems must operate without the constant information exchange that many coordination strategies typicall

Howard Lutnick's FIRST political donation since becoming Commerce Secretary was a $5 MILLION donation to House Republicans 1 month before th…

Howard Lutnick's FIRST political donation since becoming Commerce Secretary was a $5 MILLION donation to House Republicans 1 month before they interviewed him about his ties to Jeffrey Epstein. Let th

researchyann-lecun--x

Imagine2Real: Towards Zero-shot Humanoid-Object Interaction via Video Generative Priors

arXiv:2605.22272v1 Announce Type: cross Abstract: Whole-body Humanoid-Object Interaction (HOI) is bottlenecked by the scarcity of high-fidelity 3D data. While video generative priors offer a promising

Improved DDIM Sampling with Moment Matching Gaussian Mixtures

arXiv:2311.04938v5 Announce Type: replace Abstract: We propose using a Gaussian Mixture Model (GMM) as reverse transition operator (kernel) within the Denoising Diffusion Implicit Models (DDIM) framew

Improving 3D Labeling in Self-Driving by Inferring Vehicle Information using Vision Language Models

arXiv:2605.21747v1 Announce Type: new Abstract: We present an approach to improve 3D vehicle labeling in self-driving applications through zero-shot inference of vehicle information, leveraging Vehicl

Improving Viewpoint-Invariance and Temporal Consistency for Action Detection

arXiv:2605.22695v1 Announce Type: new Abstract: Viewpoint change invariance and action temporal consistency are critical aspects for the effective deployment of human action detection of untrimmed vid

In Silico Modeling of the RAMPHO Buffer: Dissociating Informational and Energetic Masking via Phonetic Entropy in Deep Neural Networks

arXiv:2605.22465v1 Announce Type: new Abstract: The fundamental challenge of listening in multi-talker environments is a cognitive bottleneck, defined by the Ease of Language Understanding (ELU) model

Industrial Dual-Arm Box Handling via Online Inertial Estimation and Convex Wrench Optimization

arXiv:2605.22021v1 Announce Type: new Abstract: Industrial robotic object handling often involves boxes and packages whose mass and center of mass are not known in advance. These uncertainties affect

Internal narratives parameterise affective states

arXiv:2502.09487v3 Announce Type: replace Abstract: Characterising how we verbalise our feelings is central to psychological assessment and intervention, yet the mapping between narrative and affectiv

Interpreting and Enhancing Emotional Circuits in Large Vision-Language Models via Cross-Modal Information Flow

arXiv:2605.21980v1 Announce Type: new Abstract: Large Vision-Language Models (LVLMs) represent a significant leap towards empathetic agents, demonstrating remarkable capabilities in emotion understand

Label tree semantic losses for rich multi-class medical image segmentation

arXiv:2507.15777v3 Announce Type: replace Abstract: Rich and accurate medical image segmentation is poised to underpin the next generation of AI-defined clinical practice by delineating critical anato

LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning

arXiv:2605.22012v1 Announce Type: new Abstract: Joint audio-visual reasoning is essential for omnimodal understanding, yet current multimodal large language models (MLLMs) still struggle when reasonin

LFX: Towards Unified Light Field Dense Semantic Segmentation and Salient Object Detection

arXiv:2503.00747v2 Announce Type: replace Abstract: Light field cameras capture multi-view observations within a single exposure. However, existing studies are typically tailored to specific LF repres

LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?

arXiv:2510.07962v2 Announce Type: replace Abstract: Large language models (LLMs) have demonstrated remarkable progress in reasoning, often through supervised fine-tuning (SFT). However, SFT is resourc

Mind the Gaps: Multi-Robot Feedback-Driven Ergodic Coverage in Unknown Environments

arXiv:2605.21719v1 Announce Type: new Abstract: In this work, we address the problem of multi-robot adaptive coverage, where teams of robots perform dynamic sampling by continuously adjusting their po

More Context, Larger Models, or Moral Knowledge? A Systematic Study of Schwartz Value Detection in Political Texts

arXiv:2605.22641v1 Announce Type: new Abstract: Detecting Schwartz values in political text is difficult because implicit cues often depend on surrounding arguments and fine-grained distinctions betwe

Motion Design for Grasp-Based Dynamic Locomotion in Microgravity

arXiv:2605.21704v1 Announce Type: new Abstract: Locomotion in microgravity often relies on sparsely and irregularly arranged anchors, motivating grasp-based mobility with multiple limbs. In this setti

MotionDPS: Motion-Compensated 3D Brain MRI Reconstruction

arXiv:2605.22121v1 Announce Type: new Abstract: Magnetic resonance imaging (MRI) is highly susceptible to patient motion due to its relatively long acquisition times and the fact that data are acquire

MRecover: A Conditional Generative Model for Recovering Motion-Corrupted MR images Using AI Generated Contrast

arXiv:2605.21669v1 Announce Type: new Abstract: Hippocampal subfield segmentation requires high-resolution T2w turbo spin echo (TSE) MRI, yet this sequence is susceptible to motion artifacts, leading

MuKV: Multi-Grained KV Cache Compression for Long Streaming Video Question-Answering

arXiv:2605.22269v1 Announce Type: new Abstract: Long streaming video QA remains challenging due to growing visual tokens and limited reasoning length of large language models (LLMs). KV-caching stores

Multi-scale interaction network for stereo image super-resolution

arXiv:2605.21913v1 Announce Type: new Abstract: Stereo image super-resolution aims to generate high-resolution images by leveraging complementary information from binocular systems. Although previous

Multi-Stage Training for Abusive Comment Detection in Indic Languages

arXiv:2605.22380v1 Announce Type: new Abstract: In recent years social media has become an increasingly popular tool for communication. People use it to share their ideas, exchange information, and di

My new article 'Toward a science of intelligence: unifying physics, neuroscience and AI' https://www.amacad.org/publication/daedalus/toward-…

My new article 'Toward a science of intelligence: unifying physics, neuroscience and AI' https://www.amacad.org/publication/daedalus/toward-science-of-intelligence-unifying-physics-neuroscience-ai pub

researchyann-lecun--x

No Pose, No Problem in 4D: Feed-Forward Dynamic Gaussians from Unposed Multi-View Videos

arXiv:2605.22190v1 Announce Type: new Abstract: Recent feed-forward 3D gaussian splatting methods have made dramatic progress on individual aspects of 3D scene reconstruction, but no existing method j

OCELOT: Odometry and Contact Estimation for Legged Robots

arXiv:2605.21863v1 Announce Type: new Abstract: One of the significant challenges in legged robotics is achieving accurate odometry using only onboard proprioceptive sensors. In this study, we present

On the Complexity of Entailment for Cumulative Propositional Dependence Logics

arXiv:2605.21113v1 Announce Type: cross Abstract: This paper establishes and proves complexity results for entailment for cumulative propositional dependence logic and for cumulative propositional log

One prompt is not enough: Instruction Sensitivity Undermines Embedding Model Evaluation

arXiv:2605.22544v1 Announce Type: new Abstract: Instruction embedding models have become common among state-of-the-art models, however are evaluated using a single prompt per task. The single-point ev

Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models

arXiv:2410.12771v2 Announce Type: replace-cross Abstract: The ability to discover new materials with desirable properties is critical for numerous applications from helping mitigate climate change to

Optical Quantum Mixed-State Reconstruction With Multiple Deep Learning Approaches

arXiv:2407.01734v4 Announce Type: replace-cross Abstract: Quantum state tomography is a crucial technique for characterizing the state of a quantum system, which is essential for many applications in

Pattern-and-root inflectional morphology: the Arabic broken plural

arXiv:2605.22310v1 Announce Type: new Abstract: We present a substantially implemented model of description of the inflectional morphology of Arabic nouns, with special attention to the management of

PIU: Proximity-guided Identity Unlearning in ID-Conditioned Diffusion Models

arXiv:2605.22311v1 Announce Type: new Abstract: Identity-conditioned diffusion models enable high-quality and identity-consistent face generation, but they also raise severe privacy concerns, as model

Planning in the LLM Era: Building for Reliability and Efficiency

arXiv:2605.21902v1 Announce Type: cross Abstract: Growing attention to intelligent agents has put a spotlight on one of their central capabilities: planning. Early attempts to leverage large language

PolycubeNet: A Dual-latent Diffusion Model for Polycube-Based Hexahedral Mesh Generation

arXiv:2605.20274v1 Announce Type: cross Abstract: Hexahedral meshes are widely used in simulation pipelines, yet automatic generation remains challenging for complex CAD geometries. Polycube-based hex

Probabilistic Attribution For Large Language Models

arXiv:2605.21726v1 Announce Type: new Abstract: The generative nature of Large Language Models (LLMs) is reflected in the conditional probabilities they compute to sample each response token given the

Proportional Selection in Networks

arXiv:2502.03545v2 Announce Type: replace-cross Abstract: We address the problem of selecting k representative nodes from a network, aiming to achieve two objectives: identifying the most influential

Quantifying Full-Body Immersion

arXiv:2605.22521v1 Announce Type: new Abstract: Humanity is at the forefront of yet another digital revolution, where the lines between real and virtual worlds are dissolving, reshaping how we perceiv

REACH: Hand Pose Estimation from Room Corners

arXiv:2605.22231v1 Announce Type: new Abstract: We introduce a novel 3D hand pose estimator that can accurately recover the shape and pose of people's hands in a room from afar, typically from fixed c

Real-Time Auto-Optimization in Unknown Environments via Structure-Exploiting Dual Control for Exploration and Exploitation

arXiv:2605.22431v1 Announce Type: new Abstract: This paper develops a fast numerical dual control for exploration and exploitation (DCEE) method to address auto-optimization problems in unknown enviro

Representability-Aware Neural Networks for Reduced Density Matrices: Application to Fractional Chern Insulators

arXiv:2605.20326v1 Announce Type: cross Abstract: We develop a representability-aware and interpolable neural network (NN) framework for predicting two-particle reduced density matrices (2-RDMs). The

Rethinking Token Reduction for Diffusion Models via Output-Similarity-Awareness

arXiv:2605.22011v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) achieve superior image generation quality but suffer from quadratic computational complexity relative to token count. Whil