NVIDIA Research

ARDY achieves real-time, controllable 3D human motion generation that adapts seamlessly to dynamic text prompts and complex kinematic constraints.

CGVQ achieves a remarkable 20% reduction in bits per pixel while maintaining visual quality, revolutionizing Gaussian-based image compression.

Audex achieves state-of-the-art audio understanding and generation while maintaining the reasoning prowess of its text-only foundation, all through a unified architecture.

Current avatar systems are more diverse than ever, yet foundational prior learning is often overlooked in discussions of photorealistic digital humans.

Current VLMs struggle with specialized domains, failing to adapt effectively in both zero-shot and ICL scenarios, revealing critical gaps in their spatio-temporal reasoning abilities.

ROSA revolutionizes robot factory operations by boosting productivity up to 12.06x through innovative shared GPU-pool serving and factory-focused scheduling.

STRATA achieves 50x better energy efficiency than conventional storm-resolving models while delivering realistic global weather simulations at kilometer-scale resolution.

Claw-like agents are vulnerable to severe security breaches, with malicious plugins achieving a 100% success rate in attacks.

Tri-serve redefines energy efficiency in multimodal inference by addressing hidden power inefficiencies, achieving a 22% boost without latency trade-offs.

ASR models can exhibit drastically different performance depending on user preferences, revealing hidden quality disparities in traditional benchmarks.

Policies trained in SimFoundry's automated environments achieve up to 40% higher success rates in real-world tasks by leveraging affordance-preserving scene variations.

Physically aligned video models can boost robotic manipulation success rates by over 50% compared to traditional methods.

Distinct model capabilities reveal that relational context significantly influences mental health assessments, with Claude-3-Haiku and GPT-4o leading in classification and trigger detection, respectively.

Intent-aware training can dramatically enhance safety classifier performance, revealing that faithful intent modeling is a powerful supervision signal.

NeuMatEx outperforms PBR techniques by extracting complex neural materials with unprecedented visual fidelity and precision from multi-view images.

SR-PPO achieves significant gains in reasoning tasks by effectively assigning credit to individual tokens from a single rollout, transforming how we approach reinforcement learning in language models.

Current multimodal conversational models miss critical emotional cues, revealing a significant gap in their ability to engage in nuanced human-like dialogue.

FAR-LIO cuts odometry latency by 38.4% while improving accuracy, setting a new standard for real-time performance in autonomous racing.

GRAFT enables robots to manipulate unseen objects with just one demonstration by leveraging geometric similarities, outperforming traditional semantic retrieval methods.

SHERLOC boosts code repair agents' effectiveness by improving fault localization accuracy while slashing token usage by over 23%.

AI data centers can now act as dynamic grid assets, reducing peak electricity demand while ensuring priority workloads remain unaffected.

Abandoning arithmetic logic for string similarity allows LLMs to achieve unprecedented accuracy in deducing logical rules from binary strings.

Achieving up to 1.90X speedup in video generation without sacrificing fidelity, ScalingAttention redefines efficiency in Diffusion Transformers.

Bagpiper-TTS can seamlessly transform natural language requests into high-quality speech across diverse applications, outperforming traditional TTS systems.

Coding agents can now autonomously refine robotic manipulation policies to achieve a staggering 99% success rate on complex tasks, revolutionizing real-world robotics.

Self-correcting models can achieve unprecedented fidelity and plausibility in generative tasks by actively learning from their own alignment errors.

Post-training on synthesized safety-critical scenarios can dramatically enhance the reliability of autonomous driving systems, reducing failures in rare but critical events.

LLMs can pass Taiwanese lawyer qualification exams but struggle with precise legal citations, revealing critical gaps in their legal reasoning capabilities.

Diffusion-Proof not only surpasses AR LLMs in theorem proving but also solves challenging problems that state-of-the-art models fail to address.

Combining learning and geometric optimization, this framework achieves a 60.9% grasp success rate, outperforming traditional methods by a significant margin.

Long-form speech generation can now achieve remarkable coherence and naturalness without the need for extensive retraining on long-form datasets.

Tail latency in LLM serving can be cut by up to 50% without relying on length predictions, reshaping how we optimize inference performance.

SR-REAL's dual-path reasoning framework allows spatial VLMs to excel in both linguistic deduction and 3D geometric inference, significantly enhancing performance on complex spatial reasoning tasks.

A single-line code change can restore diversity and fidelity in video generation models, outperforming even the original teacher models.

Tactile-reactive policies can boost robotic manipulation success rates by over 30% through innovative data collection and a new Mixture-of-Transformers architecture.

Extracting action signals from 32,041 hours of human video enables CAIP to outperform leading vision encoders in robotic manipulation tasks by over 30%.

Agents can become "addicted" to visible rewards, sacrificing safety for short-term gains, raising alarms about AI alignment in real-world applications.

cuTile Rust achieves 7 TB/s for element-wise operations on the NVIDIA B200 GPU, all while ensuring memory safety in GPU kernel programming.

SPARC reduces noisy labels by leveraging task structure, enabling robots to learn from more reliable demonstrations and outperforming traditional methods in real-world applications.

MSA slashes per-token attention compute by over 28x while maintaining competitive performance, revolutionizing how LLMs can handle ultra-long contexts.

Decoupling modality processing in VLA models leads to a staggering 95.2% success rate in complex manipulation tasks, far surpassing traditional synchronous approaches.

VIPIR achieves orders-of-magnitude higher throughput for private information retrieval while slashing communication and memory overheads, revolutionizing large-scale database privacy.

Naively scaling test-time compute is wasteful; strategically allocating it with DIRECT can enhance embodied agent performance while slashing latency by up to 65%.

VLMs trained on the new 4DP-QA dataset show marked improvements in understanding complex 4D scenes, revealing the critical role of disentangling motion dynamics.

City-scale reconstructions with over 1 billion Gaussian splats reveal a breakthrough in multi-GPU efficiency and detail, surpassing current state-of-the-art methods by more than 25 times.

DEHP dramatically boosts the success rates of high-precision robotic tasks by dynamically adjusting execution horizons based on task complexity.

VoLoAgent outperforms traditional manipulation systems by seamlessly integrating planning, monitoring, and recovery in real-time, transforming how robots handle complex tasks in dynamic environments.

Retaining the right evidence before a query can boost long-horizon agent performance by over 70% in F1 score, transforming how we think about memory management in AI.

NVSHMEM's innovative device-side symmetric-memory model could redefine GPU communication strategies, pushing the boundaries of hardware performance.

A real-time generative world model can synthesize complex driving scenarios that traditional simulators struggle to capture, enabling safer and more effective evaluation of autonomous vehicle policies.

Static evidence selection fails under budget constraints, revealing the need for adaptive strategies in retrieval-augmented systems.

Cosmos 3 sets a new benchmark for omnimodal models, outperforming existing state-of-the-art in both Text-to-Image and Image-to-Video tasks.

Achieving zero-shot generalization in robotic grasping across diverse gripper designs could revolutionize how robots interact with their environments.

Disagreement among security scanners reveals that 81.9% of flagged skills are identified by only one scanner, challenging the reliability of single-scanner assessments in agent-skill security.

Reward models trained only on success are fundamentally misaligned with human values, leading to dangerous over-rewarding of poor robot behaviors.

Steering imaginations in video world models can reveal critical failure points in robotic actions that traditional methods might overlook.

Achieving a 40x speedup in training for deformable simulations could revolutionize real-time applications in robotics and animation.

Forget scaling laws: a single looped transformer block, iterated explicitly, crushes billion-parameter feed-forward networks at multi-view 3D reconstruction.

Current vision-speech agents are surprisingly bad at mimicking the subtle, real-time audio-visual cues that make human conversation feel natural.

Stop wasting compute on redundant code generation attempts: CPPO boosts pass@$K$ by explicitly coordinating exploration across diverse algorithmic strategies.

Get up to 1.79x faster ViT inference on high-resolution images without sacrificing accuracy by surgically replacing full-attention blocks with cheaper alternatives *after* pre-training.

Masking just 5% of attention heads in vision-language models tanks performance on long-context tasks, revealing a surprisingly sparse and critical set of "multimodal retrieval heads" that attend to both text and images.

Ditch the clunky tool-use pipelines: STORM teaches video-language models to reason about space and time using *internalized* latent trajectories, slashing inference costs while boosting accuracy.

Relight 3D assets 25x faster with a feed-forward network that distills relightable representations from large reconstruction models, sidestepping expensive per-scene optimization.

Forget assuming NaNs and single-bit flips are the main culprits in GPU silent data corruption; this study reveals they're surprisingly rare, demanding a rethink of fault modeling.

Hierarchical power allocation in datacenters can achieve near-perfect satisfaction ratios, even with oversubscription, by using a novel three-phase QP/LP optimization policy.

Speculative decoding, typically used post-RL, can be integrated directly into RL training loops to accelerate LLM rollout generation by up to 2.5x.

VLN agents can navigate more accurately in zero-shot settings by "looking forward, now, and backward," mimicking human navigational strategies.

Looping language models isn't just for single agents anymore: Recursive Multi-Agent Systems (RecursiveMAS) show that agent collaboration itself can be scaled through recursion, yielding faster and more efficient problem-solving.

Near-field lighting? No problem: 8DNA pre-bakes complex light transport into neural representations, outperforming prior methods with faster inference and lower training costs.

Forget GPU-centric designs: AMMA slashes attention latency by 15x and energy consumption by 7x with a memory-centric architecture for long-context LLMs.

Multimodal models can now achieve state-of-the-art performance in real-world tasks like document understanding and audio-video comprehension with significantly reduced inference latency thanks to novel token-reduction techniques.

Forget clunky animation pipelines – MotionBricks lets you assemble real-time, high-quality character motions like LEGOs, even controlling robots.

See where your citations are coming from with a single command, thanks to CiteRadar's open-source platform that automatically generates interactive maps and detailed researcher profiles from your Google Scholar ID.

Bridging the offline-streaming gap in ASR is now more achievable: a single RNN-Transducer model can deliver high accuracy in both settings, thanks to a novel consistency regularization technique.

Squeeze up to 3.2x more performance from your long-context LLMs by intelligently splitting attention computation between CPU and GPU.

Sonata outperforms traditional models in clinical kinematic assessments, achieving better fall-risk predictions with a fraction of the parameters.

Flint enables unprecedented flexibility in exploring distributed ML design spaces by leveraging compiler insights, allowing for workload representation across any cluster size.

LLMs aren't just making us more productive, they're subtly inflating our egos by making us think we're smarter than we actually are.

LLM agents can autonomously evolve and improve a million-line EDA tool, discovering optimizations that surpass human-designed heuristics.

RoboLab exposes critical performance gaps in leading robotic models, revealing that high-fidelity simulations can better assess generalization than traditional benchmarks.

Scaling up LLMs doesn't uniformly improve context handling; instead, it paradoxically amplifies the tendency to copy irrelevant tokens while simultaneously improving resistance to misinformation.

Even moderate GPU fault rates can catastrophically derail LLM training, depending on the specific hardware datapath and numerical precision format.

Finally, a video generation model lets you puppeteer objects and their reactions independently, all while freely moving the camera.

Swap out slow, one-token-at-a-time generation in VLMs for a 6x speed boost, without sacrificing quality, using a surprisingly simple direct conversion to block-diffusion decoding.

Unlock the power of cutting-edge photon-counting CT imaging on your existing routine chest CT scans, boosting lesion detection by 10-15%.

Forget complex per-sample loss calculations – this simple three-line code injection uses batch loss smoothing to prune 20-50% of training data without sacrificing performance.

Serving both image and video diffusion models on the same hardware? GENSERVE's step-level preemption and dynamic resource allocation can boost your service level agreement (SLA) attainment by up to 44%.

GPIR shatters the PIR performance barrier, achieving 300x speedups on GPUs by rethinking kernel design and data layout to overcome memory bottlenecks exposed by multi-client batching.

Gaussian Splatting gets a high-frequency boost: Neural Harmonic Textures unlock significantly more detail in primitive-based 3D reconstructions without sacrificing speed.

Achieve 49% and 19% better Chamfer distance than state-of-the-art dynamic surface reconstruction methods on Hi4D and CMU Panoptic datasets, respectively, by enforcing temporal consistency in Gaussian Splatting.

Claims of quantum advantage in electronic structure calculations must now contend with DMRG benchmarks achieving CAS(89,102) on Fe$_5$S$_{12}$H$_4^{5-}$, pushing the boundaries of classical computation.

A 30B MoE model can now achieve Gold Medal-level performance in IMO, IOI, and ICPC, rivaling frontier models with 20x more parameters.

World Action Models can ditch the slow, iterative "imagine-then-execute" loop at test time without sacrificing performance, achieving a 4x speedup.

Current image generation unlearning methods are surprisingly brittle: adversarial image prompts, optimized with attention-guided masking, can effectively resurrect supposedly "forgotten" concepts.

Stop wrestling with incompatible human body models: SOMA lets you mix and match SMPL, SMPL-X, and more, unlocking the power of diverse datasets in a single, differentiable pipeline.

Forget expensive real-world data collection: a massive, diverse synthetic dataset enables surprisingly effective zero-shot transfer for robotic manipulation.

A hybrid cuVSLAM-based visual SLAM system achieves superior mapping accuracy in real-world logistics environments, outperforming other VO/VSLAM approaches.

Stop wasting precious GPU memory: this new cache-semantic hash table library achieves up to 3.9 billion key-value lookups per second, outperforming standard approaches by up to 9.4x.

Rivaling English's GigaSpeech in scale, TAGARELA unlocks the potential for state-of-the-art Portuguese speech models with its nearly 9,000 hours of podcast audio.