Search: “inference”

Adaptive Thinking: Large Language Models Know When to Think in Latent Space

researchapple-ml-research

Recent advances in large language models (LLMs) test-time computing have introduced the capability to perform intermediate chain-of-thought (CoT) reasoning (thinking) before generating answers. While

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Applications

Bayesian Inverse Transition Learning: Learning Dynamics From Near-Optimal Trajectories

DGX agent

arXiv:2411.05174v2 Announce Type: replace Abstract: We consider the problem of estimating the transition dynamics T^* from near-optimal expert trajectories in the context of offline model-based reinfo

applicationsarxiv-cs-lg

29 Apr 2026

Can We Change the Stroke Size for Easier Diffusion?

arXiv:2603.26783v2 Announce Type: replace Abstract: Diffusion models can be challenged in the low signal-to-noise regime, where they have to make pixel-level predictions despite the presence of high n

CodeOCR: On the Effectiveness of Vision Language Models in Code Understanding

arXiv:2602.01785v2 Announce Type: replace Abstract: Large Language Models (LLMs) have achieved remarkable success in source code understanding, yet as software systems grow in scale, computational eff

researcharxiv-cs-cl

Tutorials

Diffusion Model for Manifold Data: Score Decomposition, Curvature, and Statistical Complexity

arXiv:2603.20645v2 Announce Type: replace Abstract: Diffusion models have become a leading framework in generative modeling, yet their theoretical understanding -- especially for high-dimensional data

tutorialsarxiv-cs-lg

DouC: Dual-Branch CLIP for Training-Free Open-Vocabulary Segmentation

arXiv:2604.24997v1 Announce Type: new Abstract: Open-vocabulary semantic segmentation requires assigning pixel-level semantic labels while supporting an open and unrestricted set of categories. Traini

local-aiarxiv-cs-cv

From Cursed to Competitive: Closing the ZO-FO Gap via Input-to-State Stability

arXiv:2604.25372v1 Announce Type: cross Abstract: While it is generally understood that zeroth-order (ZO) algorithms have an extra dependency on their number of iterations for any choice of parameters

How RL Unlocks the Aha Moment in Geometric Interleaved Reasoning

arXiv:2603.01070v2 Announce Type: replace Abstract: Solving complex geometric problems inherently requires interleaved reasoning: a tight alternation between constructing diagrams and performing logic

safetyarxiv-cs-cl

Investigation into In-Context Learning Capabilities of Transformers

model-releasesarxiv-cs-lg

arXiv:2604.25858v1 Announce Type: new Abstract: Transformers have demonstrated a strong ability for in-context learning (ICL), enabling models to solve previously unseen tasks using only example input

Learning biophysical models of gene regulation with probability flow matching

arXiv:2604.25062v1 Announce Type: cross Abstract: Cellular differentiation is governed by gene regulatory networks, the high-dimensional stochastic biochemical systems that determine the transcription

Leveraging Previous-Traversal Point Cloud Map Priors for Camera-Based 3D Object Detection and Tracking

arXiv:2604.25405v1 Announce Type: new Abstract: Camera-based 3D object detection and tracking are central to autonomous driving, yet precise 3D object localization remains fundamentally constrained by

local-aiarxiv-cs-cv

Lightweight Real-Time Rendering Parameter Optimization via XGBoost-Driven Lookup Tables

model-releasesarxiv-cs-cv

arXiv:2604.25178v1 Announce Type: new Abstract: Achieving a desirable balance between rendering quality and real-time performance is a long-standing challenge in modern game and rendering engines, par

Making AI-Assisted Grant Evaluation Auditable without Exposing the Model

arXiv:2604.25200v1 Announce Type: cross Abstract: Public agencies are beginning to consider large language models (LLMs) as decision-support tools for grant evaluation. This creates a practical govern

Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling

arXiv:2604.25578v1 Announce Type: new Abstract: We present Marco-MoE, a suite of fully open multilingual sparse Mixture-of-Experts (MoE) models. Marco-MoE features a highly sparse design in which only

researcharxiv-cs-cl

MobileLLM-Flash: Latency-Guided On-Device LLM Design for Industry Scale Deployment

arXiv:2603.15954v2 Announce Type: replace Abstract: Real-time AI experiences call for on-device large language models (OD-LLMs) optimized for efficient deployment on resource-constrained hardware. The

local-aiarxiv-cs-lg

Multimodal Contextualized Support for Enhancing Video Retrieval System

arXiv:2412.07584v2 Announce Type: replace Abstract: Current video retrieval systems, especially those used in competitions, primarily focus on querying individual keyframes or images rather than encod

Mutual Forcing: Dual-Mode Self-Evolution for Fast Autoregressive Audio-Video Character Generation

arXiv:2604.25819v1 Announce Type: new Abstract: In this work, we propose Mutual Forcing, a framework for fast autoregressive audio-video generation with long-horizon audio-video synchronization. Our a

Null Measurability at the Symmetrization Interface in VC Learning

arXiv:2604.25028v1 Announce Type: new Abstract: Recent work revisiting measurability in the fundamental theorem of statistical learning imposes Borel measurability of ghost-gap suprema. We show that,

QFlash: Bridging Quantization and Memory Efficiency in Vision Transformer Attention

arXiv:2604.25306v1 Announce Type: new Abstract: FlashAttention improves efficiency through tiling, but its online softmax still relies on floating-point arithmetic for numerical stability, making full

Qwen3.6 27B on dual RTX 5060 Ti 16GB with vLLM: ~60 tok/s, 204k context working

Qwen3.6 27B is a 27-billion parameter language model that can achieve approximately 60 tokens per second throughput when running on dual RTX 5060 Ti GPUs with 16GB memory each, using the vLLM inferenc

local-air-localllama

RABC-Net: Reliability-Aware Annotation-Free Skin Lesion Segmentation for Low-Resource Dermoscopy

arXiv:2604.05594v2 Announce Type: replace Abstract: Pixel-level annotation is costly in low-resource dermoscopy. We present RABC-Net, a reliability-aware annotation-free segmentation system that combi

local-aiarxiv-cs-cv

Robust Deepfake Detection: Mitigating Spatial Attention Drift via Calibrated Complementary Ensembles

arXiv:2604.25889v1 Announce Type: new Abstract: Current deepfake detection models achieve state-of-the-art performance on pristine academic datasets but suffer severe spatial attention drift under rea

applicationsarxiv-cs-cv

SARE: Sample-wise Adaptive Reasoning for Training-free Fine-grained Visual Recognition

model-releasesarxiv-cs-cv

arXiv:2603.17729v3 Announce Type: replace Abstract: Recent advances in Large Vision-Language Models (LVLMs) have enabled training-free Fine-Grained Visual Recognition (FGVR). However, effectively expl

Self-DACE++: Robust Low-Light Enhancement via Efficient Adaptive Curve Estimation

model-releasesarxiv-cs-cv

arXiv:2604.25367v1 Announce Type: new Abstract: In this paper, we present Self-DACE++, an improved unsupervised and lightweight framework for Low-Light Image Enhancement (LLIE), building upon our prev

The founder’s AI foundation: The top announcements for startups from Next ‘26

model-releasesgoogle-cloud-ai

The momentum is undeniable: the world’s fastest-growing AI startups are building with Google Cloud. Instead of stitching together fragmented point solutions, founders are building their businesses her

The Role of Symmetry in Optimizing Overparameterized Networks

arXiv:2604.25150v1 Announce Type: new Abstract: Overparameterization is central to the success of deep learning, yet the mechanisms by which it improves optimization remain incompletely understood. We

Towards Robust Deep Learning-based Rumex Obtusifolius Detection from Drone Images

arXiv:2604.25316v1 Announce Type: new Abstract: Domain adaptation (DA) addresses the challenge of transferring a machine learning model trained on a source domain to a target domain with a different d

A Divergence-Based Method for Weighting and Averaging Model Predictions

arXiv:2604.24172v1 Announce Type: cross Abstract: This paper uses a minimum divergence framework to introduce a new way of calculating model weights that can be used to average probabilistic predictio

Tutorials

A few notes on how to get started with building LLM Knowledge Bases. @karpathy popularized it but most people don't know where to start. Eve…

A few notes on how to get started with building LLM Knowledge Bases. @karpathy popularized it but most people don't know where to start. Everyone should be creating LLM Wikis. Live session tomorrow. S

tutorialsdair-ai--x

A Lightweight Explainable Guardrail for Prompt Safety

arXiv:2602.15853v2 Announce Type: replace-cross Abstract: We propose a lightweight explainable guardrail (LEG) method to detect unsafe prompts. LEG uses a multi-task learning architecture to jointly l

safetyarxiv-cs-ai

Adversary-Free Counterfactual Prediction via Information-Regularized Representations

arXiv:2510.15479v2 Announce Type: replace Abstract: We study counterfactual prediction under assignment bias and propose a mathematically grounded, information-theoretic approach that removes treatmen

safetyarxiv-cs-lg

Always Tell Me The Odds: Fine-grained Conditional Probability Estimation

arXiv:2505.01595v2 Announce Type: replace-cross Abstract: We present a state-of-the-art model for fine-grained probability estimation of propositions conditioned on context. Recent advances in large l

researcharxiv-cs-ai

B-FIRE: Binning-Free Diffusion Implicit Neural Representation for Hyper-Accelerated Motion-Resolved MRI

arXiv:2601.06166v2 Announce Type: replace Abstract: Accelerated dynamic volumetric magnetic resonance imaging (4DMRI) is essential for applications relying on motion resolution. Existing 4DMRI produce

BEAR: Towards Beam-Search-Aware Optimization for Recommendation with Large Language Models

arXiv:2601.22925v2 Announce Type: replace-cross Abstract: Recent years have seen a rapid surge in research leveraging Large Language Models (LLMs) for recommendation. These methods typically employ su

applicationsarxiv-cs-ai

Betting for Sim-to-Real Performance Evaluation

arXiv:2604.24018v1 Announce Type: new Abstract: This paper studies the problem of robot performance evaluation, focusing on how to obtain accurate and efficient estimates of real-world behavior under

safetyarxiv-cs-ro

BitRL: Reinforcement Learning with 1-bit Quantized Language Models for Resource-Constrained Edge Deployment

model-releasesarxiv-cs-lg

arXiv:2604.24273v1 Announce Type: new Abstract: The deployment of intelligent reinforcement learning (RL) agents on resource-constrained edge devices remains a fundamental challenge due to the substan

Agents

BlindGuard: Safeguarding LLM-based Multi-Agent Systems under Unknown Attacks

arXiv:2508.08127v2 Announce Type: replace Abstract: The security of LLM-based multi-agent systems (MAS) is critically threatened by propagation vulnerability, where malicious agents can distort collec

agentsarxiv-cs-ai

Bridging Restoration and Generation Manifolds in One-Step Diffusion for Real-World Super-Resolution

arXiv:2604.24136v1 Announce Type: new Abstract: Pretrained diffusion models have revolutionized real-world image super-resolution (Real-ISR) but suffer from computational bottlenecks due to iterative

applicationsarxiv-cs-cv

BurstGP: Enhancing Raw Burst Image Super Resolution with Generative Priors

arXiv:2604.23508v1 Announce Type: new Abstract: Burst image super resolution (BISR) aims to construct a single high-resolution (HR) image by aggregating information from multiple low-resolution (LR) f

Cardiac Stability Theory: An Axiomatically Grounded Framework for Continuous Cardiac Health Monitoring via Smartphone Photoplethysmography

arXiv:2604.23876v1 Announce Type: new Abstract: We present Cardiac Stability Theory (CST), an axiomatically grounded framework formally defining cardiovascular health as a stability margin around a ca

CASP: Support-Aware Offline Policy Selection for Two-Stage Recommender Systems

arXiv:2604.23022v1 Announce Type: cross Abstract: Two-stage recommender systems first choose a candidate generator and then rank items within the generated set. Because the generator decides which ite

safetyarxiv-cs-lg

CF-VLA: Efficient Coarse-to-Fine Action Generation for Vision-Language-Action Policies

arXiv:2604.24622v1 Announce Type: cross Abstract: Flow-based vision-language-action (VLA) policies offer strong expressivity for action generation, but suffer from a fundamental inefficiency: multi-st

local-aiarxiv-cs-ai

CLIP-Guided Data Augmentation for Night-Time Image Dehazing

arXiv:2604.05500v2 Announce Type: replace Abstract: Nighttime image dehazing faces a more complex degradation pattern than its daytime counterpart, as haze scattering couples with low illumination, no

Collocation-based Robust Physics Informed Neural Networks for time-dependent simulations of pollution propagation under thermal inversion conditions on Spitsbergen

arXiv:2604.23003v1 Announce Type: new Abstract: In this paper, we propose a Physics-Informed Neural Network framework for time-dependent simulations of pollution propagation originating from moving em

applicationsarxiv-cs-lg

Decoding the mechanisms of the Hattrick football manager game using Bayesian network structure learning

arXiv:2504.09499v2 Announce Type: replace-cross Abstract: Hattrick is a free web-based probabilistic football manager game with over 200,000 users competing for titles at national and international le

researcharxiv-cs-ai

Defusing the Trigger: Plug-and-Play Defense for Backdoored LLMs via Tail-Risk Intrinsic Geometric Smoothing

model-releasesarxiv-cs-ai

arXiv:2604.24162v1 Announce Type: cross Abstract: Defending against backdoor attacks in large language models remains a critical practical challenge. Existing defenses mitigate these threats but typic

Discriminator-Guided Adaptive Diffusion for Source-Free Test-Time Adaptation under Image Corruptions

arXiv:2604.23636v1 Announce Type: new Abstract: In this work, we study Source-Free Unsupervised Domain Adaptation under corruption-induced domain shifts, where performance degradation is caused by nat

Tutorials

Doloris: Dual Conditional Diffusion Implicit Bridges with Sparsity Masking Strategy for Unpaired Single-Cell Perturbation Estimation

arXiv:2506.21107v3 Announce Type: replace Abstract: Estimating single-cell responses across various perturbations facilitates the identification of key genes and enhances drug screening, significantly

tutorialsarxiv-cs-lg