All entries

Optimal Representations for Generalized Contrastive Learning with Imbalanced Datasets

arXiv:2605.11291v1 Announce Type: new Abstract: In this paper, we provide a computable characterization of the geometry of optimal representations in Contrastive Learning (CL) when the classes are imb

Optimistic Dual Averaging Unifies Modern Optimizers

DGX agent

arXiv:2605.11172v1 Announce Type: new Abstract: We introduce SODA, a generalization of Optimistic Dual Averaging, which provides a common perspective on state-of-the-art optimizers like Muon, Lion, Ad

researcharxiv-cs-lg

13 May 2026

Optimizing 4D Wires for Sparse 3D Abstraction

arXiv:2605.11977v1 Announce Type: new Abstract: We present a unified framework for 3D geometric abstraction using a single continuous 4D wire, parameterized as a B-spline with spatial coordinates and

ORBIT: Preserving Foundational Language Capabilities in GenRetrieval via Origin-Regulated Merging

arXiv:2605.12419v1 Announce Type: new Abstract: Despite the rapid advancements in large language model (LLM) development, fine-tuning them for specific tasks often results in the catastrophic forgetti

researcharxiv-cs-cl

ORCE: Order-Aware Alignment of Verbalized Confidence in Large Language Models

arXiv:2605.12446v1 Announce Type: cross Abstract: Large language models (LLMs) often produce answers with high certainty even when they are incorrect, making reliable confidence estimation essential f

safetyarxiv-cs-cl

Oscillators Are All You Need: Irregular Time Series Modelling via Damped Harmonic Oscillators with Closed-Form Solutions

arXiv:2602.12139v2 Announce Type: replace Abstract: Transformers excel at time series modelling through attention mechanisms that capture long-term temporal patterns. However, they assume uniform time

OTT-Vid: Optimal Transport Temporal Token Compression for Video Large Language Models

arXiv:2605.11803v1 Announce Type: new Abstract: As Video Large Language Models (Video-LLMs) scale to longer and more complex videos, their inference cost grows rapidly due to the large volume of visua

OUI as a Structural Observable: Towards an Activation-Centric View of Neural Network Training

arXiv:2605.11570v1 Announce Type: new Abstract: Activation functions are what make deep networks expressive: without them, the model collapses to a linear map. Yet we still evaluate training mostly fr

Our continued commitment to Chromebooks, and looking ahead

model-releasesgoogle-cloud-ai

At the Android Show yesterday, we introduced Googlebooks: a new category of premium laptops built with Gemini’s helpfulness at the core. Designed for Gemini Intelligence, Googlebooks will give persona

Our evaluations show that frontier AI's cyber capabilities are advancing quickly. The length of cyber tasks frontier models can complete has…

Our evaluations show that frontier AI's cyber capabilities are advancing quickly. The length of cyber tasks frontier models can complete has been doubling every few months, and this rate has become fa

safetygary-marcus--x

Tutorials

Our response to the TanStack npm supply chain attack

OpenAI details its response to the TanStack “Mini Shai-Hulud” supply chain attack, outlines protections taken to secure systems and signing certificates, and explains why macOS users must update OpenA

tutorialsopenai

Output Composability of QLoRA PEFT Modules for Plug-and-Play Attribute-Controlled Text Generation

model-releasesarxiv-cs-cl

arXiv:2605.12345v1 Announce Type: new Abstract: Parameter-efficient fine-tuning (PEFT) techniques offer task-specific fine-tuning at a fraction of the cost of full fine-tuning, but require separate fi

Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

arXiv:2605.11459v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models achieve remarkable flexibility and generalization beyond classical control paradigms. However, most prevailing VLA

OverNaN: NaN-Aware Oversampling for Imbalanced Learning with Meaningful Missingness

arXiv:2605.11525v1 Announce Type: new Abstract: Missing values are routinely treated as defects to be eliminated through deletion or imputation prior to machine learning. In many applied domains, howe

Overparametrized models with posterior drift

arXiv:2506.23619v2 Announce Type: replace-cross Abstract: This paper investigates the impact of posterior drift on out-of-sample forecasting accuracy in overparametrized machine learning models. We do

Oversmoothing as Representation Degeneracy in Neural Sheaf Diffusion

arXiv:2605.11178v1 Announce Type: new Abstract: Neural Sheaf Diffusion (NSD) generalizes diffusion-based Graph Neural Networks by replacing scalar graph Laplacians with sheaf Laplacians whose learned

Overtrained, Not Misaligned

model-releasesarxiv-cs-lg

arXiv:2605.12199v1 Announce Type: new Abstract: Emergent misalignment (EM), where fine-tuning on a narrow task (like insecure code) causes broad misalignment across unrelated domains, was first demons

Overview of the MedHopQA track at BioCreative IX: track description, participation and evaluation of systems for multi-hop medical question answering

model-releasesarxiv-cs-cl

arXiv:2605.12313v1 Announce Type: new Abstract: Multi-hop question answering (QA) remains a significant challenge in the biomedical domain, requiring systems to integrate information across multiple s

p2p ai inference mesh

A peer-to-peer AI inference mesh is a decentralized network architecture where agents connect directly to discover peers and communicate through distributed protocols to share inference workloads . Su

local-air-ollama

Packed house for @hwchase17 Keynote @LangChain Interrupt

Harrison Chase delivered a keynote address at a LangChain event that drew a large audience, as indicated by the 'packed house' reference. The event appears to have been interrupted at some point durin

agentsharrison-chase--x

PairDropGS: Paired Dropout-Induced Consistency Regularization for Sparse-View Gaussian Splatting

arXiv:2605.12072v1 Announce Type: new Abstract: Dropout-based sparse-view 3D Gaussian Splatting (3DGS) methods alleviate overfitting by randomly suppressing Gaussian primitives during training. Existi

Paper: http://arxiv.org/abs/2605.06546 HF: http://huggingface.co/papers/2605.06546 Blog: http://nousresearch.com/token-superposition

This paper investigates token superposition, a phenomenon where language models can encode multiple token representations simultaneously in a single position, enabling more efficient use of model capa

researchnous-research--x

Parabolic Position Encoding: Vision-Centric, Principled, Extrapolatable, General

arXiv:2602.01418v2 Announce Type: replace Abstract: We propose Parabolic Position Encoding (PaPE), a parabola-based position encoding for vision modalities in attention-based architectures. Given a se

Parameter-Efficient Adaptation of Pre-Trained Vision Foundation Models for Active and Passive Seismic Data Denoising

arXiv:2605.10953v1 Announce Type: cross Abstract: The demand for high-resolution subsurface imaging and continuous Earth monitoring has driven rapid growth in active and passive seismic data from dens

Partial Model Sharing Improves Byzantine Resilience in Federated Conformal Prediction

arXiv:2605.11684v1 Announce Type: new Abstract: We propose a Byzantine-resilient federated conformal prediction (FCP) method that leverages partial model sharing, where only a subset of model paramete

Partition Tree: Conditional Density Estimation over General Outcome Spaces

arXiv:2602.04042v2 Announce Type: replace Abstract: We propose Partition Tree, a novel tree-based framework for conditional density estimation over general outcome spaces that supports both continuous

Partnering with @baseten is a joy! We love working alongside them to support the teams on the frontier.

Harrison Chase, the creator of LangChain, expressed positive sentiment about Baseten as a partnership, highlighting their collaborative work in supporting frontier AI teams. The post suggests that Bas

agentsharrison-chase--x

Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference

model-releasesarxiv-cs-lg

arXiv:2510.05497v5 Announce Type: replace-cross Abstract: Large-scale Mixture of Experts (MoE) Large Language Models (LLMs) have recently become the frontier open-weight models, achieving remarkable m

Applications

PayPal runs 74,000 weekly tasks in Perplexity Enterprise. Teams use it for model validation, channel performance, market trend research, com…

applicationsperplexity--x

PayPal runs 74,000 weekly tasks in Perplexity Enterprise. Teams use it for model validation, channel performance, market trend research, competitive intelligence, and product analysis. Read the custom

PD-4DGS:Progressive Decomposition of 4D Gaussian Splatting for Bandwidth-Adaptive Dynamic Scene Streaming

arXiv:2605.11427v1 Announce Type: new Abstract: 4D Gaussian Splatting (4DGS) enables high-quality dynamic novel view synthesis, yet current models remain monolithic bitstreams that clients must downlo

PENEX: AdaBoost-Inspired Neural Network Regularization

arXiv:2510.02107v4 Announce Type: replace Abstract: AdaBoost sequentially fits so-called weak learners to minimize an exponential loss, which penalizes misclassified data points more severely than oth

Persona-Conditioned Adversarial Prompting: Multi-Identity Red-Teaming for Adversarial Discovery and Mitigation

arXiv:2605.11730v1 Announce Type: new Abstract: Automated red-teaming for LLMs often discovers narrow attack slices, missing diverse real-world threats, and yielding insufficient data for safety fine-

Tutorials

PG-3DGS: Optimizing 3D Gaussian Splatting to Satisfy Physics Objectives

arXiv:2605.11266v1 Announce Type: new Abstract: Recent advances in Gaussian Splatting have enabled fast, high-fidelity 3D scene generation, yet these methods remain purely visual and lack an understan

tutorialsarxiv-cs-cv

Physics Aware Neural Networks: Denoising for Magnetic Navigation

arXiv:2602.13690v2 Announce Type: replace Abstract: Magnetic-anomaly navigation, leveraging small-scale variations in the Earth's magnetic field, is a promising alternative when GPS is unavailable or

Physics-Informed Graph Neural Networks for Frequency-Aware Optical Aberration Correction

arXiv:2512.05683v2 Announce Type: replace Abstract: Optical aberrations significantly degrade image quality in microscopy, particularly when imaging deeper into samples. These aberrations arise from d

Physics-Informed Teacher-Student Ensemble Learning for Traffic State Estimation with a Varying Speed Limit Scenario

arXiv:2605.11346v1 Announce Type: new Abstract: Physics-informed deep learning (PIDL) neural networks have shown their capability as a useful instrument for transportation practitioners in utilizing t

local-aiarxiv-cs-lg

Picasso: Holistic Scene Reconstruction with Physics-Constrained Sampling

arXiv:2602.08058v2 Announce Type: replace Abstract: In the presence of occlusions and measurement noise, geometrically accurate scene reconstructions -- which fit the sensor data -- can still be physi

Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation

arXiv:2605.12492v1 Announce Type: new Abstract: We introduce Pion, a spectrum-preserving optimizer for large language model (LLM) training based on orthogonal equivalence transformation. Unlike additi

PIVOT: Bridging Planning and Execution in LLM Agents via Trajectory Refinement

arXiv:2605.11225v1 Announce Type: cross Abstract: Large language model (LLM)-based agents frequently generate seemingly coherent plans that fail upon execution due to infeasible actions, constraint vi

agentsarxiv-cs-lg

PointCaM: Cut-and-Mix for Open-Set Point Cloud Learning

arXiv:2212.02011v3 Announce Type: replace Abstract: Point cloud learning is receiving increasing attention. However, most existing point cloud models lack the practical ability to deal with the unavoi

PointForward: Feedforward Driving Reconstruction through Point-Aligned Representations

arXiv:2605.11594v1 Announce Type: new Abstract: High-fidelity reconstruction of driving scenes is crucial for autonomous driving. While recent feedforward 3D Gaussian Splatting (3DGS) methods enable f

agentsarxiv-cs-cv

PointGS: Semantic-Consistent Unsupervised 3D Point Cloud Segmentation with 3D Gaussian Splatting

arXiv:2605.11520v1 Announce Type: new Abstract: Unsupervised point cloud segmentation is critical for embodied artificial intelligence and autonomous driving, as it mitigates the prohibitive cost of d

Poolside is hosting a 2-day model research hackathon in London. Join us to push an open-weight agent model as far as you can. RL and fine-tu…

Poolside is hosting a 2-day model research hackathon in London. Join us to push an open-weight agent model as far as you can. RL and fine-tune Laguna XS.2, our latest-generation model, on Prime Intell

local-aiclem-delangue--x

POP: Prior-Fitted First-Order Optimization Policies

model-releasesarxiv-cs-lg

arXiv:2602.15473v2 Announce Type: replace Abstract: Gradient-based optimizers are highly sensitive to design choices in their adaptive learning rate mechanisms. To address this limitation, we introduc

PoseBridge: Bridging the Skeletonization Gap for Zero-Shot Skeleton-Based Action Recognition

arXiv:2605.11497v1 Announce Type: new Abstract: Zero-shot skeleton-based action recognition (ZSSAR) is typically treated as a skeleton-text alignment problem: encode joint-coordinate sequences, align

PoseCompass: Intelligent Synthetic Pose Selection for Visual Localization

arXiv:2605.12144v1 Announce Type: new Abstract: In visual localization, Absolute Pose Regression (APR) enables real-time 6-DoF camera pose inference from single images, yet critically depends on fine-

local-aiarxiv-cs-cv

Position: Universal Aesthetic Alignment Narrows Artistic Expression

arXiv:2512.11883v3 Announce Type: replace-cross Abstract: Over-aligning image generation models to a generalized aesthetic preference conflicts with user intent, particularly when 'anti-aesthetic' out

Post-ADC Inference: Valid Inference After Active Data Collection

arXiv:2605.11511v1 Announce Type: cross Abstract: The validity of statistical inference depends critically on how data are collected. When data gathered through active data collection (ADC) are reused