Agglomerative models have recently emerged as a powerful approach to training vision foundation models, leveraging multi-teacher distillation from existing models such as CLIP, DINO, and SAM. This strategy enables the efficient creation of robust models, combining the strengths of individual teachers while significantly reducing computational and resource demands. In this paper, we thoroughly analyze state-of-the-art agglomerative models, identifying critical challenges including resolution mode shifts, teacher imbalance, idiosyncratic teacher artifacts, and an excessive number of output tokens. To address these issues, we propose several novel solutions: multi-resolution training, mosaic augmentation, and improved balancing of teacher loss functions. Specifically, in the context of Vision Language Models, we introduce a token compression technique to maintain high-resolution information within a fixed token count. We release our top-performing variants at multiple scales (-B, -L, -H, and -g), along with inference code and pretrained weights.

Visual language models (VLMs) have made significant advances in accuracy in recent years. However, their efficiency has received much less attention. This paper introduces NVILA, a family of open VLMs designed to optimize both efficiency and accuracy. Building on top of VILA, we improve its model architecture by first scaling up the spatial and temporal resolutions, and then compressing visual tokens. This "scale-then-compress" approach enables NVILA to efficiently process high-resolution images and long videos. We also conduct a systematic investigation to enhance the efficiency of NVILA throughout its entire lifecycle, from training to deployment. NVILA matches or surpasses the accuracy of many leading open and proprietary VLMs across a wide range of image and video benchmarks. At the same time, it reduces training costs by 1.9-5.1×, prefilling latency by 1.6-2.2×, and decoding latency by 1.2-2.8×.

Generalist vision language models (VLMs) have made significant strides in computer vision, but they fall short in specialized fields like healthcare, where expert knowledge is essential. Current large multimodal models like Gemini and GPT-4o are insufficient for medical tasks due to their reliance on memorized internet knowledge rather than the nuanced expertise required in healthcare. Meanwhile, existing medical VLMs (e.g. Med-Gemini) often lack expert consultation as part of their design, and many rely on outdated, static datasets that were not created with modern, large deep learning models in mind. VLMs are usually trained in three stages: vision pre-training, vision-language pre-training, and instruction fine-tuning (IFT). IFT has been typically applied using a mixture of generic and healthcare data. In contrast, we propose that for medical VLMs, a fourth stage of specialized IFT is necessary, which focuses on medical data and includes information from domain expert models. Domain expert models developed for medical use are crucial because they are specifically trained for certain clinical tasks, e.g. to detect tumors and classify abnormalities through segmentation and classification, which learn fine-grained features of medical data−features that are often too intricate for a VLM to capture effectively. This paper introduces a new framework, VILA-M3, for medical VLMs that utilizes domain knowledge via expert models. We argue that generic VLM architectures alone are not viable for real-world clinical applications and on-demand usage of domain-specialized expert model knowledge is critical for advancing AI in healthcare. Through our experiments, we show an improved state-of-the-art (SOTA) performance with an average improvement of ~9% over the prior SOTA model Med-Gemini and ~6% over models trained on the specific tasks. Our approach emphasizes the importance of domain expertise in creating precise, reliable VLMs for medical applications.

High-resolution perception of visual details is crucial for daily tasks. Current vision pre-training, however, is still limited to low resolutions (e.g., 378×378 pixels) due to the quadratic cost of processing larger images. We introduce PS3 that scales CLIP-style vision pre-training to 4K resolution with a near-constant cost. Instead of contrastive learning on global image representation, PS3 is pre-trained by selectively processing local regions and contrasting them with local detailed captions, enabling high-resolution representation learning with greatly reduced computational overhead. The pre-trained PS3 is able to both encode the global image at low resolution and selectively process local high-resolution regions based on their saliency or relevance to a text prompt. When applying PS3 to multi-modal LLM (MLLM), the resulting model, named VILA-HD, significantly improves high-resolution visual perception compared to baselines without high-resolution vision pre-training such as AnyRes and S2while using up to 4.3× fewer tokens. PS3 also unlocks appealing scaling properties of VILA-HD, including scaling up resolution for free and scaling up test-time compute for better performance. Compared to state of the arts, VILA-HD outperforms previous MLLMs such as NVILA and Qwen2-VL across multiple benchmarks and achieves better efficiency than latest token pruning approaches. Finally, we find current benchmarks do not require 4K-resolution perception, which motivates us to propose 4KPro, a new benchmark of image QA at 4K resolution, on which VILA-HD outperforms all previous MLLMs, including a 14.5% improvement over GPT-4o, and a 3.2% improvement and 2.96× speedup over Qwen2-VL.

Vision-language-action models (VLAs) have shown potential in leveraging pretrained vision-language models and diverse robot demonstrations for learning generalizable sensorimotor control. While this paradigm effectively utilizes large-scale data from both robotic and non-robotic sources, current VLAs primarily focus on direct input–output mappings, lacking the intermediate reasoning steps crucial for complex manipulation tasks. As a result, existing VLAs lack temporal planning or reasoning capabilities. In this paper, we introduce a method that incorporates explicit visual chain-of-thought (CoT) reasoning into vision-language-action models (VLAs) by predicting future image frames autoregressively as visual goals before generating a short action sequence to achieve these goals. We introduce CoT-VLA, a state-of-the-art 7B VLA that can understand and generate visual and action tokens. Our experimental results demonstrate that CoT-VLA achieves strong performance, outperforming the state-of-the-art VLA model by 17% in real-world manipulation tasks and 6% in simulation benchmarks. Videos are available at: https://cot-vla.github.io/.

Vision Language Models (VLMs) have rapidly advanced in integrating visual and textual reasoning, powering applications across high-resolution image understanding, long-video analysis, and multi-turn conversation. However, their scalability remains limited by the growing number of visual tokens that dominate inference latency. We present SparseVILA, a new paradigm for efficient VLM inference that decouples visual sparsity across the prefilling and decoding stages. SparseVILA distributes sparsity across stages by pruning redundant visual tokens during prefill and retrieving only query-relevant tokens during decoding. This decoupled design matches leading prefill pruning methods while preserving multi-turn fidelity by retaining most of the visual cache so that query-aware tokens can be retrieved at each conversation round. Built on an AWQ-optimized inference pipeline, SparseVILA achieves up to 4.0 times faster prefilling, 2.5 times faster decoding, and an overall 2.6 times end-to-end speedup on long-context video tasks -- while improving accuracy on document-understanding and reasoning tasks. By decoupling query-agnostic pruning and query-aware retrieval, SparseVILA establishes a new direction for efficient multimodal inference, offering a training-free, architecture-agnostic framework for accelerating large VLMs without sacrificing capability.

We introduce DC-AR, a novel masked autoregressive (AR) text-to-image generation framework that delivers superior image generation quality with exceptional computational efficiency. Due to the tokenizers' limitations, prior masked AR models have lagged behind diffusion models in terms of quality or efficiency. We overcome this limitation by introducing DC-HT - a deep compression hybrid tokenizer for AR models that achieves a 32x spatial compression ratio while maintaining high reconstruction fidelity and cross-resolution generalization ability. Building upon DC-HT, we extend MaskGIT and create a new hybrid masked autoregressive image generation framework that first produces the structural elements through discrete tokens and then applies refinements via residual tokens. DC-AR achieves state-of-the-art results with a gFID of 5.49 on MJHQ-30K and an overall score of 0.69 on GenEval, while offering 1.5-7.9x higher throughput and 2.0-3.5x lower latency compared to prior leading diffusion and autoregressive models.

Long-context capability is critical for multi-modal foundation models, especially for long video understanding. We introduce LongVILA, a full-stack solution for long-context visual-language models by co-designing the algorithm and system. For model training, we upgrade existing VLMs to support long video understanding by incorporating two additional stages, i.e., long context extension and long video supervised fine-tuning. However, training on long video is computationally and memory intensive. We introduce the long-context Multi-Modal Sequence Parallelism (MM-SP) system that efficiently parallelizes long video training and inference, enabling 2M context length training on 256 GPUs without any gradient checkpointing. LongVILA efficiently extends the number of video frames of VILA from 8 to 2048, achieving 99.8% accuracy in 6,000-frame (more than 1 million tokens) video needle-in-a-haystack. LongVILA-7B demonstrates strong accuracy on 9 popular video benchmarks, e.g., 65.1% VideoMME with subtitle. Besides, MM-SP is 2.1x - 5.7x faster than ring style sequence parallelism and 1.1x - 1.4x faster than Megatron with a hybrid context and tensor parallelism. Moreover, it seamlessly integrates with Hugging Face Transformers.

We introduce Hybrid Autoregressive Transformer (HART), the first autoregressive (AR) visual generation model capable of directly generating 1024x1024 images, rivaling diffusion models in image generation quality. Existing AR models face limitations due to the poor image reconstruction quality of their discrete tokenizers and the prohibitive training costs associated with generating 1024px images. To address these challenges, we present the hybrid tokenizer, which decomposes the continuous latents from the autoencoder into two components: discrete tokens representing the big picture and continuous tokens representing the residual components that cannot be represented by the discrete tokens. The discrete component is modeled by a scalable-resolution discrete AR model, while the continuous component is learned with a lightweight residual diffusion module with only 37M parameters. Compared with the discrete-only VAR tokenizer, our hybrid approach improves reconstruction FID from 2.11 to 0.30 on MJHQ-30K, leading to a 31% generation FID improvement from 7.85 to 5.38. HART also outperforms state-of-the-art diffusion models in both FID and CLIP score, with 4.5-7.7$\times$ higher throughput and 6.9-13.4$\times$ lower MACs. Our code is open sourced at https://github.com/mit-han-lab/hart.

VILA-U is a Unified foundation model that integrates Video, Image, Language understanding and generation. Traditional visual language models (VLMs) use separate modules for understanding and generating visual content, which can lead to misalignment and increased complexity. In contrast, VILA-U employs a single autoregressive next-token prediction framework for both tasks, eliminating the need for additional components like diffusion models. This approach not only simplifies the model but also achieves near state-of-the-art performance in visual language understanding and generation. The success of VILA-U is attributed to two main factors: the unified vision tower that aligns discrete visual tokens with textual inputs during pretraining, which enhances visual perception, and autoregressive image generation can achieve similar quality as diffusion models with high-quality dataset. This allows VILA-U to perform comparably to more complex models using a fully token-based autoregressive framework.

FP8 training has emerged as a promising method for improving training efficiency. Existing frameworks accelerate training by applying FP8 computation to linear layers while leaving optimizer states and activations in higher precision, which fails to fully optimize memory usage. This paper introduces COAT (**C**ompressing **O**ptimizer States and **A**ctivations for FP8 **T**raining), a novel FP8 training framework designed to significantly reduce memory footprint when training large models. COAT addresses current limitations through two key innovations: (1) **Dynamic Range Expansion**, which aligns optimizer state distributions more closely with the FP8 representation range, thereby reducing quantization error, and (2) **Mixed-Granularity Activation Quantization**, which optimizes activation memory using a combination of per-tensor and per-group quantization strategies. Experiments demonstrate that COAT effectively reduces end-to-end training memory footprint by **1.54×** compared to BF16 while achieving nearly lossless performance across various tasks, such as Large Language Model pretraining and fine-tuning and Vision Language Model training. COAT also achieves a **1.43×** end-to-end training speedup compared to BF16, performing on par with or surpassing TransformerEngine's speedup. COAT enables efficient full-parameter training of large models on fewer GPUs, and facilitates doubling the batch size in distributed training settings, providing a practical solution for scaling large-scale model training. Code will be released upon publication.

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096$\times$4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8$\times$, we trained an AE that can compress images 32$\times$, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024$\times$1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released upon publication.

We introduce a full-stack framework that scales up reasoning in vision-language models (VLMs) to long videos, leveraging reinforcement learning. We address the unique challenges of long video reasoning by integrating three critical components: (1) a large-scale dataset, LongVideo-Reason, comprising 104K long video QA pairs with high-quality reasoning annotations across diverse domains such as sports, games, and vlogs; (2) a two-stage training pipeline that extends VLMs with chain-of-thought supervised fine-tuning (CoT-SFT) and reinforcement learning (RL); and (3) a training infrastructure for long video RL, named Multi-modal Reinforcement Sequence Parallelism (MR-SP), which incorporates sequence parallelism and a vLLM-based engine tailored for long video, using cached video embeddings for efficient rollout and prefilling. In our experiments, LongVILA-R1-7B achieves strong performance on video benchmarks, reaching 65.1% and 71.1% accuracy on VideoMME without and with subtitles, respectively, and consistently outperforming LongVILA-7B across multiple benchmarks. Moreover, LongVILA-R1-7B supports processing up to 8,192 video frames per video, and configurable FPS settings. Notably, our MR-SP system achieves up to 2.1x speedup on long video RL training. In addition, we release our training system for public availability that supports RL training on various modalities (video, text, and audio), various models (VILA and Qwen series), and even image and video generation models. On a single A100 node (8 GPUs), it supports RL training on hour-long videos (e.g., 3,600 frames). Code and models are available at https://github.com/NVlabs/Long-RL

Video generation models have rapidly progressed, positioning themselves as video world models capable of supporting decision-making applications like robotics and autonomous driving. However, current benchmarks fail to rigorously evaluate these claims, focusing only on general video quality, ignoring important factors to world models such as physics adherence.To bridge this gap, we propose WorldModelBench, a benchmark designed to evaluate the world modeling capabilities of video generation models in application-driven domains. WorldModelBench offers two key advantages: (1) Against to nuanced world modeling violations: By incorporating instruction-following and physics-adherence dimensions, WorldModelBench detects subtle violations, such as irregular changes in object size that breach the mass conservation law—issues overlooked by prior benchmarks. (2) Aligned with large-scale human preferences: We crowd-source 67K human labels to accurately measure 14 frontier models. Using our high-quality human labels, we further fine-tune an accurate judger to automate the evaluation procedure, achieving 9.9% lower error in predicting world modeling violations than GPT-4o with 2B parameters. In addition, we demonstrate that training to align human annotations by maximizing the rewards from the judger noticeably improve the world modeling capability. The dataset is hosted in HuggingFace at https://huggingface.co/datasets/Efficient-Large-Model/worldmodelbench. The code to run evaluation is available at https://github.com/WorldModelBench-Team/WorldModelBench.

Generalization remains one of the most important desiderata for robust robot learning systems. While recently proposed approaches show promise in generalization to novel objects, semantic concepts, or visual distribution shifts, generalization to new tasks remains challenging. For example, a language-conditioned policy trained on pick-and-place tasks will not be able to generalize to a folding task, even if the arm trajectory of folding is similar to pick-and-place. Our key insight is that this kind of generalization becomes feasible if we represent the task through rough trajectory sketches. We propose a policy conditioning method using such rough trajectory sketches, which we call RT-Trajectory, that is practical, easy to specify, and allows the policy to effectively perform new tasks that would otherwise be challenging to perform. We find that trajectory sketches strike a balance between being detailed enough to express low-level motion-centric guidance while being coarse enough to allow the learned policy to interpret the trajectory sketch in the context of situational visual observations. In addition, we show how trajectory sketches can provide a useful interface to communicate with robotic policies -- they can be specified through simple human inputs like drawings or videos, or through automated methods such as modern image-generating or waypoint-generating methods. We evaluate RT-Trajectory at scale on a variety of real-world robotic tasks, and find that RT-Trajectory is able to perform a wider range of tasks compared to language-conditioned and goal-conditioned policies, when provided the same training data.

We present Self-Adaptive Robust Attention for Robotics Transformers (SARA-RT): a new paradigm for addressing the emerging challenge of scaling up Robotics Transformers (RT) for on-robot deployment. SARA-RT relies on the new method of fine-tuning proposed by us, called up-training. It converts pre-trained or already fine-tuned Transformer-based robotic policies of quadratic time complexity (including massive billion-parameter vision-language-action models or VLAs), into their efficient linear-attention counterparts maintaining high quality. We demonstrate the effectiveness of SARA-RT by speeding up: (a) the class of recently introduced RT-2 models [1], the first VLA robotic policies pre-trained on internet-scale data, as well as (b) Point Cloud Transformer (PCT) robotic policies operating on large point clouds. We complement our results with the rigorous mathematical analysis providing deeper insight into the phenomenon of SARA.

We present a scalable, bottom-up and intrinsically diverse data collection scheme that can be used for high-level reasoning with long and medium horizons and that has 2.2x higher throughput compared to traditional narrow top-down step-by-step collection. We collect realistic data by performing any user requests within the entirety of 3 office buildings and using multiple embodiments (robot, human, human with grasping tool). With this data, we show that models trained on all embodiments perform better than ones trained on the robot data only, even when evaluated solely on robot episodes. We explore the economics of collection costs and find that for a fixed budget it is beneficial to take advantage of the cheaper human collection along with robot collection. We release a large and highly diverse (29,520 unique instructions) dataset dubbed RoboVQA containing 829,502 (video, text) pairs for robotics-focused visual question answering. We also demonstrate how evaluating real robot experiments with an intervention mechanism enables performing tasks to completion, making it deployable with human oversight even if imperfect while also providing a single performance metric. We demonstrate a single video-conditioned model named RoboVQA-VideoCoCa trained on our dataset that is capable of performing a variety of grounded high-level reasoning tasks in broad realistic settings with a cognitive intervention rate 46% lower than the zeroshot state of the art visual language model (VLM) baseline and is able to guide real robots through long-horizon tasks. The performance gap with zero-shot state-of-the-art models indicates that a lot of grounded data remains to be collected for real-world deployment, emphasizing the critical need for scalable data collection approaches. Finally, we show that video VLMs significantly outperform single-image VLMs with an average error rate reduction of 19% across all VQA tasks. Thanks to video conditioning and dataset diversity, the model can be used as general video value functions (e.g. success and affordance) in situations where actions needs to be recognized rather than states, expanding capabilities and environment understanding for robots. Data and videos are available at robovqa.github.io

We propose Token Turing Machines (TTM), a sequential, autoregressive Transformer model with memory for real-world sequential visual understanding. Our model is inspired by the seminal Neural Turing Machine, and has an external memory consisting of a set of tokens which summarise the previous history (i.e., frames). This memory is efficiently addressed, read and written using a Transformer as the processing unit/controller at each step. The model's memory module ensures that a new observation will only be processed with the contents of the memory (and not the entire history), meaning that it can efficiently process long sequences with a bounded computational cost at each step. We show that TTM outperforms other alternatives, such as other Transformer models designed for long sequences and recurrent neural networks, on two real-world sequential visual understanding tasks: online temporal activity detection from videos and vision-based robot action policy learning. Code is publicly available at: https://github.com/google-research/scenic/tree/main/scenic/projects/token.turing.

Reinforcement learning (RL) provides a theoretical framework for continuously improving an agent’s behavior via trial and error. However, efficiently learning policies from scratch can be very difficult, particularly for tasks that present exploration challenges. In such settings, it might be desirable to initialize RL with an existing policy, offline data, or demonstrations. However, naively performing such initialization in RL often works poorly, especially for value-based methods. In this paper, we present a meta algorithm that can use offline data, demonstrations, or a pre-existing policy to initialize an RL policy, and is compatible with any RL approach. In particular, we propose Jump-Start Reinforcement Learning (JSRL), an algorithm that employs two policies to solve tasks: a guide-policy, and an exploration-policy. By using the guide-policy to form a curriculum of starting states for the exploration-policy, we are able to efficiently improve performance on a set of simulated robotic tasks. We show via experiments that it is able to significantly outperform existing imitation and reinforcement learning algorithms, particularly in the small-data regime. In addition, we provide an upper bound on the sample complexity of JSRL and show that with the help of a guide-policy, one can improve the sample complexity for non-optimism exploration methods from exponential in horizon to polynomial.

We propose a framework to enable multipurpose assistive mobile robots to autonomously wipe tables to clean spills and crumbs. This problem is challenging, as it requires planning wiping actions while reasoning over uncertain latent dynamics of crumbs and spills captured via high-dimensional visual observations. Simultaneously, we must guarantee constraints satisfaction to enable safe deployment in unstructured cluttered environments. To tackle this problem, we first propose a stochastic differential equation to model crumbs and spill dynamics and absorption with a robot wiper. Using this model, we train a vision-based policy for planning wiping actions in simulation using reinforcement learning (RL). To enable zero-shot sim-to-real deployment, we dovetail the RL policy with a whole-body trajectory optimization framework to compute base and arm joint trajectories that execute the desired wiping motions while guaranteeing constraints satisfaction. We extensively validate our approach in simulation and on hardware. Video of experiments: https://youtu.be/inORKP4F3EI

Simulation is an essential tool to develop and benchmark autonomous vehicle planning software in a safe and cost-effective manner. However, realistic simulation requires accurate modeling of multi-agent interactive behaviors to be trustworthy, behaviors which can be highly nuanced and complex. To address these challenges, we introduce Waymax, a new data-driven simulator for autonomous driving in multi-agent scenes, designed for large-scale simulation and testing. Waymax uses publicly-released, real-world driving data (e.g., the Waymo Open Motion Dataset) to initialize or play back a diverse set of multi-agent simulated scenarios. It runs entirely on hardware accelerators such as TPUs/GPUs and supports in-graph simulation for training, making it suitable for modern large-scale, distributed machine learning workflows. To support online training and evaluation, Waymax includes several learned and hard-coded behavior models that allow for realistic interaction within simulation. To supplement Waymax, we benchmark a suite of popular imitation and reinforcement learning algorithms with ablation studies on different design decisions, where we highlight the effectiveness of routes as guidance for planning agents and the ability of RL to overfit against simulated agents.

Recent progress in large language models (LLMs) has demonstrated the ability to learn and leverage Internet-scale knowledge through pre-training with autoregressive models. Unfortunately, applying such models to settings with embodied agents, such as robots, is challenging due to their lack of experience with the physical world, inability to parse non-language observations, and ignorance of rewards or safety constraints that robots may require. On the other hand, language-conditioned robotic policies that learn from interaction data can provide the necessary grounding that allows the agent to be correctly situated in the real world, but such policies are limited by the lack of high-level semantic understanding due to the limited breadth of the interaction data available for training them. Thus, if we want to make use of the semantic knowledge in a language model while still situating it in an embodied setting, we must construct an action sequence that is both likely according to the language model and also realizable according to grounded models of the environment. We frame this as a problem similar to probabilistic filtering: decode a sequence that both has high probability under the language model and high probability under a set of grounded model objectives. We demonstrate how such grounded models can be obtained across three simulation and real-world domains, and that the proposed decoding strategy is able to solve complex, long-horizon embodiment tasks in a robotic setting by leveraging the knowledge of both models.

Many computer vision tasks address the problem of scene understanding and are naturally interrelated e.g. object classification, detection, scene segmentation, depth estimation, etc. We show that we can leverage the inherent relationships among collections of tasks, as they are trained jointly, supervising each other through their known relationships via consistency losses. Furthermore, explicitly utilizing the relationships between tasks allows improving their performance while dramatically reducing the need for labeled data, and allows training with additional unsupervised or simulated data. We demonstrate a distributed joint training algorithm with task-level parallelism, which affords a high degree of asynchronicity and robustness. This allows learning across multiple tasks, or with large amounts of input data, at scale. We demonstrate our framework on subsets of the following collection of tasks: depth and normal prediction, semantic segmentation, 3D motion and egomotion estimation, and object tracking and 3D detection in point clouds. We observe improved performance across these tasks, especially in the low-label regime.

We consider the problem of learning useful robotic skills from previously collected offline data without access to manually specified rewards or additional online exploration, a setting that is becoming increasingly important for scaling robot learning by reusing past robotic data. In particular, we propose the objective of learning a functional understanding of the environment by learning to reach any goal state in a given dataset. We employ goal-conditioned Q-learning with hindsight relabeling and develop several techniques that enable training in a particularly challenging offline setting. We find that our method can operate on high-dimensional camera images and learn a variety of skills on real robots that generalize to previously unseen scenes and objects. We also show that our method can learn to reach long-horizon goals across multiple episodes through goal chaining, and learn rich representations that can help with downstream tasks through pre-training or auxiliary objectives.

We propose a vision-based architecture search algorithm for robot manipulation learning, which discovers interactions between low dimension action inputs and high dimensional visual inputs. Our approach automatically designs architectures while training on the task – discovering novel ways of combining and attending image feature representations with actions as well as features from previous layers. The obtained new architectures demonstrate better task success rates, in some cases with a large margin, compared to a recent high performing baseline. Our real robot experiments also confirm that it improves grasping performance by 6%. This is the first approach to demonstrate a successful neural architecture search and attention connectivity search for a real-robot task.

Federated Learning is an emerging direction in distributed machine learning that en-ables jointly training a model without sharing the data. Since the data is distributed across many edge devices through wireless / long-distance connections, federated learning suffers from inevitable high communication latency. However, the latency issues are undermined in the current literature [15] and existing approaches suchas FedAvg [27] become less efficient when the latency increases. To over comethe problem, we propose \textbf{D}elayed \textbf{G}radient \textbf{A}veraging (DGA), which delays the averaging step to improve efficiency and allows local computation in parallel tocommunication. We theoretically prove that DGA attains a similar convergence rate as FedAvg, and empirically show that our algorithm can tolerate high network latency without compromising accuracy. Specifically, we benchmark the training speed on various vision (CIFAR, ImageNet) and language tasks (Shakespeare),with both IID and non-IID partitions, and show DGA can bring 2.55$\times$ to 4.07$\times$ speedup. Moreover, we built a 16-node Raspberry Pi cluster and show that DGA can consistently speed up real-world federated learning applications.

In this article, we explore the possibility of enabling cloud-based virtual space applications for better computational scalability and easy access from any end device, including future lightweight wireless head-mounted displays. In particular, we investigate virtual space applications such as virtual classroom and virtual gallery, in which the scenes and activities are rendered in the cloud, with multiple views captured and streamed to each end device. A key challenge is the high bandwidth requirement to stream all the user views, leading to high operational cost and potential large delay in a bandwidth-restricted wireless network. We propose a novel hybrid-cast approach to save bandwidth in a multi-user streaming scenario. We identify and broadcast the common pixels shared by multiple users, while unicasting the residual pixels for each user. We formulate the problem of minimizing the total bitrate needed to transmit the user views using hybrid-casting and describe our approach. A common view extraction approach and a smart grouping algorithm are proposed and developed to achieve our hybrid-cast approach. Simulation results show that the hybrid-cast approach can significantly reduce total bitrate by up to 55% and avoid congestion-related latency, compared to traditional cloud-based approach of transmitting all the views as individual unicast streams, hence addressing the bandwidth challenges of the cloud, with additional benefits in cost and delay.

Triggered by several head-mounted display (HMD) devices that have come to the market recently, such as Oculus Rift, HTC Vive, and Samsung Gear VR, significant interest has developed in virtual reality (VR) systems, experiences and applications. However, the current HMD devices are still very heavy and large, negatively affecting user experience. Moreover, current VR approaches perform rendering locally either on a mobile device tethered to an HMD, or on a computer/console tethered to the HMD. In this paper, we discuss how to enable a truly portable and mobile VR experience, with light weight VR glasses wirelessly connecting with edge/cloud computing devices that perform the rendering remotely. We investigate the challenges associated with enabling the new wireless VR approach with edge/cloud computing with different application scenarios that we implement. Specifically, we analyze the challenging bitrate and latency requirements to enable wireless VR, and investigate several possible solutions.

Motivation: We introduce PRINCESS, a privacy-preserving international collaboration framework for analyzing rare disease genetic data that are distributed across different continents. PRINCESS leverages Software Guard Extensions (SGX) and hardware for trustworthy computation. Unlike a traditional international collaboration model, where individual-level patient DNA are physically centralized at a single site, PRINCESS performs a secure and distributed computation over encrypted data, fulfilling institutional policies and regulations for protected health information. Results: To demonstrate PRINCESS' performance and feasibility, we conducted a family-based allelic association study for Kawasaki Disease, with data hosted in three different continents. The experimental results show that PRINCESS provides secure and accurate analyses much faster than alternative solutions, such as homomorphic encryption and garbled circuits (over 40 000× faster). Availability and Implementation: https://github.com/achenfengb/PRINCESS_opensource. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Although wavefront parallel processing (WPP) proposed in the HEVC standard and various inter frame WPP algorithms can achieve comparatively high parallelism, their scalability for its parallelism is still very limited due to various dependencies introduced in spatial and temporal prediction in HEVC. In this paper, we propose three types of 3 Dimensional WPP (3D-WPP) algorithms that can significantly improve the parallelism, while achieving good tradeoffs between implementation complexity, determinism, and rate-distortion (RD) performance. Experimental results show that the proposed algorithms can lead to up to 2.8 × speed up compared with existing inter frame WPP methods. While the Simple 3D-WPP and Static 3D-, WPP algorithm may introduce an BD rate loss between 0 to 4.9% as compared with existing algorithms, the more complex Dynamic 3D-WPP algorithm achieves better parallelism with virtually no coding performance loss.

In this paper, we explore the possibility of enabling cloud-based virtual classroom applications providing the advantages of computational scalability and access from any end device. In particular, we investigate a virtual classroom application in which the classroom including teacher, students and activities are rendered on the cloud, with each student view captured and streamed to students' end devices. By identifying that many student views may share common pixels, we design a novel hyper-cast system so that common pixels can be transmitted by a single broadcast stream while the residual pixels for individual student view can be transmitted by unicast in cellular networks. The hyper-cast approach can significantly reduce total bitrate needed, and therefore decrease cloud cost and cellular bandwidth. Simulation results show that the proposed hyper-cast technique can save up to 44.6% bitrate compared to traditional cloud-based approach.

In this paper, we propose an Avatar based Virtual Reality user training system that efficiently trains users in performing a variety of activities using a pre-recorded avatar. To evaluate and monitor the user's adherence to the avatar's instructions, the system compares the user's motion data against the avatar's motion data, with the latter established as the ground truth dataset. Unfortunately, human reaction delay may cause the motion sequences between the user and the avatar to be misaligned. Consequently, to enable accurate comparison, we analyze four signal processing time delay estimation methods—an existing method and three proposed methods—to align the motion sequences between the user and pre-recorded avatar, allowing the correct frames to be compared. Our experiments demonstrate that the proposed methods perform better data alignment than the existing method and the fourth method, which employs a novel spatial-temporal segmentation algorithm, has the highest potential to be the optimal delay estimation approach. Further, to provide real-time guidance to the user, we determine a unique tolerance threshold for each activity such that a user accuracy value below the threshold value prompts real-time guidance to correct the user and an accuracy value above the threshold is tolerated. We perform an experiment with the assistance of a physical trainer and use the experimental data to design a histogram-based method using Bayesian decision theory to determine the threshold values.

The growing popularity of auto-stereoscopic 3D displays for mobile devices, together with ubiquitous wireless networks, have fueled an increasing user expectation for rich 3D mobile multimedia experiences, including 3D display gaming. However, rendering 3D games on mobile devices requires high computational power and battery and thus may restrict users from enjoying true 3D experience for a long time. In this paper, we explore the possibility of developing Cloud Mobile 3D Display Gaming, where the 3D video rendering and encoding are performed on cloud servers, with the resulting 3D video streamed wirelessly to mobile devices with 3D displays. However, with the significantly higher bitrate requirement for 3D video, ensuring user experience may be a challenge considering the bandwidth constraints and fluctuations of mobile networks. In this paper, we propose a novel asymmetric Object of Interest (OOI) rendering approach, which adapts the rendering richness of different objects according to their importance in order to reduce the video encoding bitrate needed while maintaining a satisfactory video quality, thereby making it easier to transmit the 3D game video over wireless network. Specifically, we first develop a model to quantitatively measure the user experience by different OOI rendering settings. We also develop a model to relate the bitrate of the resulting video with the changes of different OOI Rendering settings. We further propose an optimization algorithm which uses the above two models to automatically decide the optimal OOI rendering settings for left view and right view to ensure the best user experience given the network conditions. Experiments conducted using real 4G-LTE network profiles on commercial cloud service demonstrate the improvement in user experience when the proposed optimization algorithm is applied.

Cloud gaming allows games to be rendered on the cloud server and allows the rendered videos to be encoded and streamed in real time to the player's devices. Compared with other video streaming applications, cloud gaming offers a unique opportunity to enhance the video encoding process by exploiting rendering information. In this paper, we propose two techniques to improve cloud gaming video encoding, aiming at enhancing the perceived video quality and reducing the computational complexity, respectively. First, we develop a rendering-based prioritized encoding technique to improve the perceived game video quality according to network bandwidth constraints. We first propose a technique to generate a macroblock (MB)-level saliency map for every game video frame using rendering information. Furthermore, based on such a saliency map, a prioritized rate allocation scheme is proposed to dynamically adjust the value of quantization parameter of each MB. The experimental results indicate that the perceptual quality can be greatly improved using the proposed technique. We also develop a rendering-based encoding acceleration technique that utilizes rendering information to reduce the computational complexity of video encoding. This technique mainly consists of two parts. First, we propose a method to directly calculate the motion vectors (MVs) without employing the compute intensive motion search procedure. Second, based on the computed MVs, we propose a fast mode selection algorithm to reduce the number of candidate modes of each MB. The experimental results show that the proposed technique can achieve more than 42% saving in encoding time with very limited degradation in video quality.

The emerging High Efficiency Video Coding (HEVC) standard achieves significant performance improvement over H.264/AVC standard at a cost of much higher complexity. In this paper, we propose a ultra fast H.264/AVC to HEVC transcoder for multi-core processors implementing Wave front Parallel Processing (WPP) and SIMD acceleration, along with expedited motion estimation (ME) and mode decision (MD) by utilizing information extracted from the input H.264/AVC stream. Experiments using standard HEVC test bit streams show that the proposed transcoder achieves 70x speed up over the HEVC HM 8.1 reference software (including H.264 encoding) at very small rate distortion (RD) performance loss.

In this paper, we present an improved algorithm for decoding live streamed or pre-encoded video bit streams with time-varying qualities. The algorithm extracts information available to the decoder from a high visual quality segment of the clip that has already been received and decoded, but was encoded independently from the current segment. The proposed decoder is capable of significantly improve the Quality of Experience of the user without incurring significant overhead to the storage and computational complexities of both the encoder and the decoder. We present simulation results using the HEVC reference encoder and standard test clips, and discuss areas of improvements to the algorithm and potential ways of incorporating the technique to a video streaming system or standards.

In this letter, we present a new algorithm for a single image super-resolution using the analysis sparse prior in thelαβ color space. Experimental results show that our algorithm outperforms other existing state-of-the-art methods. In addition, due to the high scalability of our algorithm, key modules of the proposed algorithm can be integrated with other super resolution algorithms.

We describe a novel and effective video enhancement algorithm for low lighting video. The algorithm works by first inverting an input low-lighting video and then applying an optimized image de-haze algorithm on the inverted video. To facilitate faster computation, temporal correlations between subsequent frames are utilized to expedite the calculation of key algorithm parameters. Simulation results show excellent enhancement results and 4× speed up as compared with the frame-wise enhancement algorithms.