Meta AI (FAIR)

A proactive memory agent can significantly enhance decision-making in long-horizon tasks by preventing critical information from being forgotten.

Automated view scheduling in SceneFrom3D transforms the landscape of outdoor 3D scene generation, enabling unprecedented control over object appearance and geometry.

Stochastic attention isn't just a regularizer; it fundamentally transforms how visual information is learned in VLMs, leading to more stable and reliable models.

Clinically structured rank allocation in BiRG-LoRA boosts medical question answering accuracy while reducing trainable parameters by over 28%.

LUNA achieves realistic 3D human animation from 2D inputs without the limitations of traditional skinning methods, enabling unprecedented flexibility and expressivity.

TriageRA-CCF reveals that leveraging source-side clinical signals can significantly enhance the performance of medical LLMs by optimizing rank budgeting dynamically.

Transformers can generate complex triangulations that are vital for advancing our understanding of Calabi-Yau manifolds in string theory.

Cross-modal prediction in MJEPA not only boosts performance but also reveals that a unified encoder can outperform traditional modality-specific approaches by leveraging shared information.

FiCA generates photorealistic avatars from a single image, achieving unprecedented visual quality and identity fidelity without the need for individual optimization.

Selective teacher intervention in multi-turn training can boost agent performance by over 13% by mitigating the impact of early errors.

Discriminator-Guided RL achieves a remarkable reduction in FID from 9.38 to 2.62, showcasing a new way to align model outputs with real data without human preferences.

RepFusion reveals that multimodal large language models can dramatically enhance denoising in text-to-image systems, outperforming traditional denoising methods.

Surprisingly, the error in estimating counts for hierarchical prefixes remains constant regardless of hierarchy height or the number of heavy hitters.

A new phase diagram reveals that cross-modal training can be actively harmful in certain contexts, guiding practitioners to choose the right approach before training.

Standard MCQA metrics can mislead evaluations by over 2 points due to phrasing sensitivity, but ParaEval cuts this gap to under 1 point, revealing true model capabilities.

Augmenting fMRI datasets with synthetic data can yield a staggering 68% boost in image retrieval accuracy, challenging traditional limits of brain decoding.

Chunk-level semantic verification in OmniOPD yields a +28.64% boost in math performance over traditional OPD, challenging the reliance on token-level logit matching.

Feedback Distillation boosts reasoning model performance by enhancing trajectory diversity and policy entropy, outperforming traditional methods like GRPO.

Jointly training a speech encoder and language model on mel-spectrograms not only boosts zero-shot speech translation, but also fixes annoying speech synthesis quirks like endless silences.

Backpropagation's gradients, while predictive of high-level visual cortex activity, march to a different hierarchical beat than the brain itself, challenging its status as a biologically plausible learning mechanism.

Extrapolating between code-generating RL agents trained on different unit test coverages unlocks better correctness-efficiency trade-offs than any single agent alone.

Classical SfM can get stuck, and feedforward reconstruction can be brittle, but combining them creates a system that's both robust and accurate.

LLMs can make ad recommendations more stable and predictable, ensuring that minor creative changes don't lead to wild swings in ad delivery.

Skip the costly data collection for new eye-tracking devices: GazePrior synthesizes realistic training data by learning a 3D prior of human eyes, enabling zero-shot transfer.

Camera pose, largely ignored in video LLMs, unlocks significant gains in spatial reasoning and even improves general video QA when used as a lightweight supervisory signal.

Controllable 3D generation takes a leap forward with 3D-ReGen, a framework that leverages an initial 3D shape for tasks like enhancement and editing, outperforming existing methods.

Human motion generation gets a dose of reality: IAM shows that explicitly modeling body morphology and identity leads to more realistic and consistent movements.

Storing user interaction histories in a normalized, immutable tier and reconstructing sequences just-in-time slashes data infrastructure costs and unlocks the potential of ultra-long sequence DLRMs.

Hallucinating LLMs in enterprise workflows can be tamed by a new Hybrid Utility Minimum Bayes Risk (HUMBR) framework that synthesizes semantic and lexical signals to achieve consensus without ground truth.

Serverless functions can get a 37% density boost and significantly reduced overhead by offloading I/O to a shared backend, without sacrificing ecosystem compatibility.

Training 3D avatar diffusion models on millions of in-the-wild videos is now possible, thanks to a clever 3D tokenization and visibility-aware training strategy that overcomes partial observability.

Scaling robot learning with human data isn't a simple "more is better" equation; alignment with robot learning objectives is key.

AI could provide a new lens on the structure of mathematics, potentially answering the age-old question of whether it is discovered or invented.

Music-grounded video editing can now produce significantly more coherent timelines thanks to a novel global-local coordination mechanism that resolves cross-segment conflicts.

Forget scaling laws: a large VLM strategically paired with a smaller model's reasoning tokens can rival the performance of a much larger, monolithic model.

Real-world coding benchmarks reveal that AI coding agents succeed more often when they iteratively validate their work with tests and static analysis, suggesting a path to better agents in unfamiliar codebases.

Stop avatars from looking like they're having a seizure: this method uses autoregressive prediction of appearance latents to create temporally stable and high-fidelity 3D Gaussian avatars.

On-policy reward modeling with LLM judges not only unlocks significant performance gains on complex mathematical reasoning tasks, but also generalizes to improve performance on simpler numerical and multiple-choice benchmarks.

Mimicking human cognition, FLAIR lets dialogue models "think while listening," boosting performance without adding latency.

Ditch the data augmentation and decoders: R2-Dreamer's Barlow Twins-inspired objective delivers faster, more versatile MBRL, especially when spotting the small stuff matters.

Elastic-Sketch's performance hinges on stream characteristics and eviction thresholds, but this work cracks the code to near-optimal configuration by deriving closed-form expressions for its limiting behavior under stationary random streams.

OmniSONAR halves cross-lingual search error on FLORES and reduces error by 15x on BIBLE, proving that truly universal sentence embeddings across thousands of languages and modalities are now within reach.

Forget scaling laws: a specialized 8B parameter translation model can outperform a 70B general-purpose LLM on 1,600 languages.

Current AI's hunger for curated data may be solved by a new architecture inspired by human cognition that flexibly switches between observation, active behavior, and meta-control.

The optimal spectrogram configuration for audio and speech analysis hinges on a nuanced interplay between front-end feature representation and back-end classifier architecture, varying significantly across tasks.

Forget exotic attention mechanisms – MobileLLM-Flash achieves up to 1.8x faster LLM prefill on mobile CPUs by smartly pruning and adapting existing architectures for on-device use.

Self-supervised video models can now learn dense features rivaling supervised methods, unlocking a 20-point jump in robot grasping success.

Forget imbalanced LoRA usage: ReMix leverages reinforcement learning to route effectively among LoRAs, boosting performance in parameter-efficient fine-tuning.

LLMs struggle to generate diverse and specific connections between concepts, even with high token budgets and "thinking" prompts, revealing a gap in creative associative reasoning.

LLM reasoning research is inadvertently paving a dangerous path towards AI situational awareness and strategic deception, demanding a re-evaluation of current safety measures.

Even the best open-weight LLMs still fail on nearly two-thirds of questions requiring reasoning over scientific tables, highlighting a persistent "execution bottleneck" in translating strategy to action.

Recursive self-improvement can boost performance by 18% in code and 17% in reasoning, but only if you can keep it from going off the rails – SAHOO provides the guardrails.

Pre-normalization in Transformers is the culprit behind the mysterious link between massive activation outliers and attention sinks, but decoupling them reveals their distinct functions: global parameterization vs. local attention modulation.

Datacenter networks are haunted by "ghosts"—topology knowledge failures due to link flaps that occur every 48 seconds at 2025 cluster scale—and existing mitigations are insufficient, but Open Atomic Ethernet offers a potential exorcism.

LLMs can ace math problems while reasoning like a drunk toddler, with 82% of correct answers arising from unstable, inconsistent logic.

SSL models can be backdoored with nearly undetectable triggers that still achieve high attack success rates, even against common defenses.

Stop leaking your face to generative AI apps: PRIVATEEDIT lets you edit images while keeping biometric data on your device.

By disentangling camera-space estimation from world-space refinement via dual diffusion models, DuoMo achieves state-of-the-art human motion reconstruction from noisy video, bypassing the limitations of parametric models.

Safety classifiers for LLMs can catastrophically fail with even minuscule embedding drift, creating dangerous blind spots in deployed safety architectures.

AI agents can now learn durable skills instead of constantly "reinventing the wheel," thanks to SkillNet's infrastructure for creating, evaluating, and connecting AI skills at scale.

Randomly throwing distortions at your watermarking model during training? Meta-FC shows meta-learning a better way, boosting robustness by up to 4.71% against combined distortions.

Unlocking the secrets of viral video ads: a new MLLM framework reveals which initial moments hook viewers and drive conversions.

A deterministic client selection method leveraging gradient updates can boost federated learning accuracy by nearly 50% in heterogeneous environments.

A global consensus on AI safety risks and capabilities has emerged from a panel of 100+ independent experts, representing a landmark effort in international collaboration.

By surgically intervening in MLLM decoding, this work cuts hallucination rates without sacrificing descriptive quality, a feat prior methods struggled to achieve.

Escape the bottleneck of translating product intent into ranking system hypotheses: GEARS offers an agentic framework that autonomously discovers and validates superior ranking policies.

Autonomous coding agents derail 30% of the time, but a lightweight intervention system can recover 90% of those misbehaviors with a single nudge.

Forget quadratic complexity: ULTRA-HSTU achieves 21x faster inference and 4-8% better engagement in large-scale recommendation by co-designing input sequences, sparse attention, and model topology.

Achieve zero-collision embedding tables in production recommenders without sacrificing training speed, unlocking better personalization via fresher and higher-quality item embeddings.

Despite growing interest, queer NLP research remains largely reactive, highlighting biases instead of building proactive solutions, leaving significant opportunities for stakeholder-driven and intersectional approaches.

Ditch ANN search altogether: MFLI learns a hierarchical index alongside item embeddings, boosting recall by up to 11.8% and cold-content delivery by 57.29% in large-scale recommender systems.

Unlock superhuman visual reasoning in multimodal models by simply giving them the ability to think step-by-step at test time.

Object-level masking in world models unlocks a 20% boost in counterfactual reasoning and drastically reduces planning costs, hinting at a path toward more efficient and robust AI agents.

Forget noisy user data: synthetic data unlocks predictable scaling laws for LLMs in recommendation, boosting recall by 130%.

Forget huge models: parameter-efficient fine-tuning turns tiny language models into code-generating powerhouses that outperform larger, untuned counterparts.

Achieve state-of-the-art UAV detection by swapping transformers for Mamba, yielding a faster and more accurate multimodal detector.

Even when trained on suboptimal data, a Bayesian in-context RL agent can achieve near-optimal decisions on unseen tasks by fusing a learned Q-value prior with in-context information and employing an upper-confidence bound for exploration.

LLM safety guardrails are far less robust than benchmarks suggest, with accuracy dropping by as much as 57% on novel adversarial attacks, and some even generating harmful content in a "helpful mode" jailbreak.

This work integrates small-molecule high-throughput screening with a deep-learning-based virtual screening approach to uncover new antibacterial compounds, illustrating a 90-fold improved hit rate over the high-throughput screening experiment used for training.

Forget scaling laws: this work shows you can get SOTA reasoning from sub-billion parameter models with *less* data, if you're smart about curation and resampling.

Dramatically slash the carbon footprint of AI training without sacrificing performance by co-designing hardware and software for modern GPUs.

Achieve up to 39.6% FLOP reduction in LLM inference without retraining or architectural changes using QuickSilver's dynamic token-level optimizations.

Ditch the pre-trained models: PAST directly learns speech tokens from phonetic data, outperforming existing methods in representation and reconstruction.

LLMs can generate plain language summaries of scientific research that are as good as human-written ones, but easier to read.

Edit the bassline, drums, or other instruments of any song with this new open-source multi-stem music generation model.