Gaussian Splatting has enabled real-time 3D human avatars with unprecedented levels of visual quality. While previous methods require a desktop GPU for real-time inference of a single avatar, we aim to squeeze multiple Gaussian avatars onto a portable virtual reality headset with real-time drivable inference. We begin by training a previous work, Animatable Gaussians, on a high quality dataset captured with 512 cameras. The Gaussians are animated by controlling base set of Gaussians with linear blend skinning (LBS) motion and then further adjusting the Gaussians with a neural network decoder to correct their appearance. When deploying the model on a Meta Quest 3 VR headset, we find two major computational bottlenecks: the decoder and the rendering. To accelerate the decoder, we train the Gaussians in UV-space instead of pixel-space, and we distill the decoder to a single neural network layer. Further, we discover that neighborhoods of Gaussians can share a single corrective from the decoder, which provides an additional speedup. To accelerate the rendering, we develop a custom pipeline in Vulkan that runs on the mobile GPU. Putting it all together, we run 3 Gaussian avatars concurrently at 72 FPS on a VR headset. Demo videos are at https://forresti.github.io/squeezeme.
Creating a high-fidelity, animatable 3D full-body avatar from a single image is a challenging task due to the diverse appearance and poses of humans and the limited availability of high-quality training data. To achieve fast and high-quality human reconstruction, this work rethinks the task from the perspectives of dataset, model, and representation. First, we introduce a large-scale HUman-centric GEnerated dataset, HuGe100K, consisting of 100K diverse, photorealistic sets of human images. Each set contains 24-view frames in specific human poses, generated using a pose-controllable image-to-multi-view model. Next, leveraging the diversity in views, poses, and appearances within HuGe100K, we develop a scalable feed-forward transformer model to predict a 3D human Gaussian representation in a uniform space from a given human image. This model is trained to disentangle human pose, body shape, clothing geometry, and texture. The estimated Gaussians can be animated without post-processing. We conduct comprehensive experiments to validate the effectiveness of the proposed dataset and method. Our model demonstrates the ability to efficiently reconstruct photorealistic humans at 1K resolution from a single input image using a single GPU instantly. Additionally, it seamlessly supports various applications, as well as shape and texture editing tasks. Project page: https://yiyuzhuang.github.io/IDOL/.
Utilizing longer contexts is increasingly essential to power better AI systems. However, the cost of attending to long contexts is high due to the involved softmax computation. While the scaled dot-product attention (SDPA) exhibits token sparsity, with only a few pivotal tokens significantly contributing to attention, leveraging this sparsity effectively remains an open challenge. Previous methods either suffer from model degradation or require considerable additional resources. We propose HashAttention --a principled approach casting pivotal token identification as a recommendation problem. Given a query, HashAttention encodes keys and queries in Hamming space capturing the required semantic similarity using learned mapping functions. HashAttention efficiently identifies pivotal tokens for a given query in this Hamming space using bitwise operations, and only these pivotal tokens are used for attention computation, significantly improving overall attention efficiency. HashAttention can reduce the number of tokens used by a factor of $1/32\times$ for the Llama-3.1-8B model with LongBench, keeping average quality loss within 0.6 points, while using only 32 bits per token auxiliary memory. At $32\times$ sparsity, HashAttention is $3{-}6\times$ faster than LightLLM and $2.5{-}4.5\times$ faster than gpt-fast on Nvidia-L4 GPU.
Procedural Content Generation (PCG) is powerful in creating high-quality 3D contents, yet controlling it to produce desired shapes is difficult and often requires extensive parameter tuning. Inverse Procedural Content Generation aims to automatically find the best parameters under the input condition. However, existing sampling-based and neural network-based methods still suffer from numerous sample iterations or limited controllability. In this work, we present DI-PCG, a novel and efficient method for Inverse PCG from general image conditions. At its core is a lightweight diffusion transformer model, where PCG parameters are directly treated as the denoising target and the observed images as conditions to control parameter generation. DI-PCG is efficient and effective. With only 7.6M network parameters and 30 GPU hours to train, it demonstrates superior performance in recovering parameters accurately, and generalizing well to in-the-wild images. Quantitative and qualitative experiment results validate the effectiveness of DI-PCG in inverse PCG and image-to-3D generation tasks. DI-PCG offers a promising approach for efficient inverse PCG and represents a valuable exploration step towards a 3D generation path that models how to construct a 3D asset using parametric models.
High-quality benchmarks are the foundation for embodied AI research, enabling significant advancements in long-horizon navigation, manipulation and rearrangement tasks. However, as frontier tasks in robotics get more advanced, they require faster simulation speed, more intricate test environments, and larger demonstration datasets. To this end, we present MS-HAB, a holistic benchmark for low-level manipulation and in-home object rearrangement. First, we provide a GPU-accelerated implementation of the Home Assistant Benchmark (HAB). We support realistic low-level control and achieve over 3x the speed of prior magical grasp implementations at a fraction of the GPU memory usage. Second, we train extensive reinforcement learning (RL) and imitation learning (IL) baselines for future work to compare against. Finally, we develop a rule-based trajectory filtering system to sample specific demonstrations from our RL policies which match predefined criteria for robot behavior and safety. Combining demonstration filtering with our fast environments enables efficient, controlled data generation at scale.
SocialED is a comprehensive, open-source Python library designed to support social event detection (SED) tasks, integrating 19 detection algorithms and 14 diverse datasets. It provides a unified API with detailed documentation, offering researchers and practitioners a complete solution for event detection in social media. The library is designed with modularity in mind, allowing users to easily adapt and extend components for various use cases. SocialED supports a wide range of preprocessing techniques, such as graph construction and tokenization, and includes standardized interfaces for training models and making predictions. By integrating popular deep learning frameworks, SocialED ensures high efficiency and scalability across both CPU and GPU environments. The library is built adhering to high code quality standards, including unit testing, continuous integration, and code coverage, ensuring that SocialED delivers robust, maintainable software. SocialED is publicly available at \url{https://github.com/RingBDStack/SocialED} and can be installed via PyPI.
In the quest to overcome language barriers, encoder-decoder models like NLLB have expanded machine translation to rare languages, with some models (e.g., NLLB 1.3B) even trainable on a single GPU. While general-purpose LLMs perform well in translation, open LLMs prove highly competitive when fine-tuned for specific tasks involving unknown corpora. We introduce LYRA (Language verY Rare for All), a novel approach that combines open LLM fine-tuning, retrieval-augmented generation (RAG), and transfer learning from related high-resource languages. This study is exclusively focused on single-GPU training to facilitate ease of adoption. Our study focuses on two-way translation between French and Mon\'egasque, a rare language unsupported by existing translation tools due to limited corpus availability. Our results demonstrate LYRA's effectiveness, frequently surpassing and consistently matching state-of-the-art encoder-decoder models in rare language translation.
This paper presents a comprehensive evaluation of state-of-the-art object detection models, including YOLOv9, YOLOv10, and RT-DETR, for the task of weed detection in smart-spraying applications focusing on three classes: Sugarbeet, Monocot, and Dicot. The performance of these models is compared based on mean Average Precision (mAP) scores and inference times on different GPU and CPU devices. We consider various model variations, such as nano, small, medium, large alongside different image resolutions (320px, 480px, 640px, 800px, 960px). The results highlight the trade-offs between inference time and detection accuracy, providing valuable insights for selecting the most suitable model for real-time weed detection. This study aims to guide the development of efficient and effective smart spraying systems, enhancing agricultural productivity through precise weed management.
Large Language Models (LLMs) have demonstrated remarkable performance across diverse tasks yet still are vulnerable to external threats, particularly LLM Denial-of-Service (LLM-DoS) attacks. Specifically, LLM-DoS attacks aim to exhaust computational resources and block services. However, existing studies predominantly focus on white-box attacks, leaving black-box scenarios underexplored. In this paper, we introduce Auto-Generation for LLM-DoS (AutoDoS) attack, an automated algorithm designed for black-box LLMs. AutoDoS constructs the DoS Attack Tree and expands the node coverage to achieve effectiveness under black-box conditions. By transferability-driven iterative optimization, AutoDoS could work across different models in one prompt. Furthermore, we reveal that embedding the Length Trojan allows AutoDoS to bypass existing defenses more effectively. Experimental results show that AutoDoS significantly amplifies service response latency by over 250$\times\uparrow$, leading to severe resource consumption in terms of GPU utilization and memory usage. Our work provides a new perspective on LLM-DoS attacks and security defenses. Our code is available at https://github.com/shuita2333/AutoDoS.
Graph neural networks (GNNs) have seen extensive application in domains such as social networks, bioinformatics, and recommendation systems. However, the irregularity and sparsity of graph data challenge traditional computing methods, which are insufficient to meet the performance demands of GNNs. Recent research has explored parallel acceleration using CUDA Cores and Tensor Cores, but significant challenges persist: (1) kernel fusion leads to false high utilization, failing to treat CUDA and Tensor Cores as independent resources, and (2) heterogeneous cores have distinct computation preferences, causing inefficiencies. To address these issues, this paper proposes FTC-GNN, a novel acceleration framework that efficiently utilizes CUDA and Tensor Cores for GNN computation. FTC-GNN introduces (1) a collaborative design that enables the parallel utilization of CUDA and Tensor Cores and (2) a sparse-to-dense transformation strategy that assigns dense matrix operations to Tensor Cores while leveraging CUDA Cores for data management and sparse edge processing. This design optimizes GPU resource utilization and improves computational efficiency. Experimental results demonstrate the effectiveness of FTC-GNN using GCN and AGNN models across various datasets. For GCN, FTC-GNN achieves speedups of 4.90x, 7.10x, and 1.17x compared to DGL, PyG, and TC-GNN, respectively. For AGNN, it achieves speedups of 5.32x, 2.92x, and 1.02x, establishing its superiority in accelerating GNN computations.
Weather forecasting plays a vital role in today's society, from agriculture and logistics to predicting the output of renewable energies, and preparing for extreme weather events. Deep learning weather forecasting models trained with the next state prediction objective on ERA5 have shown great success compared to numerical global circulation models. However, for a wide range of applications, being able to provide representative samples from the distribution of possible future weather states is critical. In this paper, we propose a methodology to leverage deterministic weather models in the design of probabilistic weather models, leading to improved performance and reduced computing costs. We first introduce \textbf{ArchesWeather}, a transformer-based deterministic model that improves upon Pangu-Weather by removing overrestrictive inductive priors. We then design a probabilistic weather model called \textbf{ArchesWeatherGen} based on flow matching, a modern variant of diffusion models, that is trained to project ArchesWeather's predictions to the distribution of ERA5 weather states. ArchesWeatherGen is a true stochastic emulator of ERA5 and surpasses IFS ENS and NeuralGCM on all WeatherBench headline variables (except for NeuralGCM's geopotential). Our work also aims to democratize the use of deterministic and generative machine learning models in weather forecasting research, with academic computing resources. All models are trained at 1.5{\deg} resolution, with a training budget of $\sim$9 V100 days for ArchesWeather and $\sim$45 V100 days for ArchesWeatherGen. For inference, ArchesWeatherGen generates 15-day weather trajectories at a rate of 1 minute per ensemble member on a A100 GPU card. To make our work fully reproducible, our code and models are open source, including the complete pipeline for data preparation, training, and evaluation, at https://github.com/INRIA/geoarches .
Traditional rule-based cybersecurity systems have proven highly effective against known malware threats. However, they face challenges in detecting novel threats. To address this issue, emerging cybersecurity systems are incorporating AI techniques, specifically deep-learning algorithms, to enhance their ability to detect incidents, analyze alerts, and respond to events. While these techniques offer a promising approach to combating dynamic security threats, they often require significant computational resources. Therefore, frameworks that incorporate AI-based cybersecurity mechanisms need to support the use of GPUs to ensure optimal performance. Many cybersecurity framework vendors do not provide sufficiently detailed information about their implementation, making it difficult to assess the techniques employed and their effectiveness. This study aims to overcome this limitation by providing an overview of the most used cybersecurity frameworks that utilize AI techniques, specifically focusing on frameworks that provide comprehensive information about their implementation. Our primary objective is to identify the deep-learning techniques employed by these frameworks and evaluate their support for GPU acceleration. We have identified a total of \emph{two} deep-learning algorithms that are utilized by \emph{three} out of 38 selected cybersecurity frameworks. Our findings aim to assist in selecting open-source cybersecurity frameworks for future research and assessing any discrepancies between deep-learning techniques used in theory and practice.
Simulation offers unique values for both enumeration and extrapolation purposes, and is becoming increasingly important for managing the massive machine learning (ML) clusters and large-scale distributed training jobs. In this paper, we build Echo to tackle three key challenges in large-scale training simulation: (1) tracing the runtime training workloads at each device in an ex-situ fashion so we can use a single device to obtain the actual execution graphs of 1K-GPU training, (2) accurately estimating the collective communication without high overheads of discrete-event based network simulation, and (3) accounting for the interference-induced computation slowdown from overlapping communication and computation kernels on the same device. Echo delivers on average 8% error in training step -- roughly 3x lower than state-of-the-art simulators -- for GPT-175B on a 96-GPU H800 cluster with 3D parallelism on Megatron-LM under 2 minutes.
This paper addresses supervised deep metric learning for open-set image retrieval, focusing on three key aspects: the loss function, mixup regularization, and model initialization. In deep metric learning, optimizing the retrieval evaluation metric, recall@k, via gradient descent is desirable but challenging due to its non-differentiable nature. To overcome this, we propose a differentiable surrogate loss that is computed on large batches, nearly equivalent to the entire training set. This computationally intensive process is made feasible through an implementation that bypasses the GPU memory limitations. Additionally, we introduce an efficient mixup regularization technique that operates on pairwise scalar similarities, effectively increasing the batch size even further. The training process is further enhanced by initializing the vision encoder using foundational models, which are pre-trained on large-scale datasets. Through a systematic study of these components, we demonstrate that their synergy enables large models to nearly solve popular benchmarks.
This work explores the YOLOv6 object detection model in depth, concentrating on its design framework, optimization techniques, and detection capabilities. YOLOv6's core elements consist of the EfficientRep Backbone for robust feature extraction and the Rep-PAN Neck for seamless feature aggregation, ensuring high-performance object detection. Evaluated on the COCO dataset, YOLOv6-N achieves 37.5\% AP at 1187 FPS on an NVIDIA Tesla T4 GPU. YOLOv6-S reaches 45.0\% AP at 484 FPS, outperforming models like PPYOLOE-S, YOLOv5-S, YOLOX-S, and YOLOv8-S in the same class. Moreover, YOLOv6-M and YOLOv6-L also show better accuracy (50.0\% and 52.8\%) while maintaining comparable inference speeds to other detectors. With an upgraded backbone and neck structure, YOLOv6-L6 delivers cutting-edge accuracy in real-time.
Recently, with the emergence of recent Multimodal Large Language Model (MLLM) technology, it has become possible to exploit its video understanding capability on different classification tasks. In practice, we face the difficulty of huge requirements for GPU resource if we need to deploy MLLMs online. In this paper, we propose COEF-VQ, a novel cascaded MLLM framework for better video quality understanding on TikTok. To this end, we first propose a MLLM fusing all visual, textual and audio signals, and then develop a cascade framework with a lightweight model as pre-filtering stage and MLLM as fine-consideration stage, significantly reducing the need for GPU resource, while retaining the performance demonstrated solely by MLLM. To demonstrate the effectiveness of COEF-VQ, we deployed this new framework onto the video management platform (VMP) at TikTok, and performed a series of detailed experiments on two in-house tasks related to video quality understanding. We show that COEF-VQ leads to substantial performance gains with limit resource consumption in these two tasks.
Large Language Models (LLMs) have become essential tools in natural language processing, finding large usage in chatbots such as ChatGPT and Gemini, and are a central area of research. A particular area of interest includes designing hardware specialized for these AI applications, with one such example being the neural processing unit (NPU). In 2023, Intel released the Intel Core Ultra processor with codename Meteor Lake, featuring a CPU, GPU, and NPU system-on-chip. However, official software support for the NPU through Intel's OpenVINO framework is limited to static model inference. The dynamic nature of autoregressive token generation in LLMs is therefore not supported out of the box. To address this shortcoming, we present NITRO (NPU Inference for Transformers Optimization), a Python-based framework built on top of OpenVINO to support text and chat generation on NPUs. In this paper, we discuss in detail the key modifications made to the transformer architecture to enable inference, some performance benchmarks, and future steps towards improving the package. The code repository for NITRO can be found here: https://github.com/abdelfattah-lab/nitro.
Effective congestion management along signalized corridors is essential for improving productivity and reducing costs, with arterial travel time serving as a key performance metric. Traditional approaches, such as Coordinated Signal Timing and Adaptive Traffic Control Systems, often lack scalability and generalizability across diverse urban layouts. We propose Fusion-based Dynamic Graph Neural Networks (FDGNN), a structured framework for simultaneous modeling of travel time distributions in both directions along arterial corridors. FDGNN utilizes attentional graph convolution on dynamic, bidirectional graphs and integrates fusion techniques to capture evolving spatiotemporal traffic dynamics. The framework is trained on extensive hours of simulation data and utilizes GPU computation to ensure scalability. The results demonstrate that our framework can efficiently and accurately model travel time as a normal distribution on arterial roads leveraging a unique dynamic graph representation of corridor traffic states. This representation integrates sequential traffic signal timing plans, local driving behaviors, temporal turning movement counts, and ingress traffic volumes, even when aggregated over intervals as short as a single cycle length. The results demonstrate resilience to effective traffic variations, including cycle lengths, green time percentages, traffic density, and counterfactual routes. Results further confirm its stability under varying conditions at different intersections. This framework supports dynamic signal timing, enhances congestion management, and improves travel time reliability in real-world applications.
Executing precise and agile flight maneuvers is critical for quadrotors in various applications. Traditional quadrotor control approaches are limited by their reliance on flat trajectories or time-consuming optimization, which restricts their flexibility. Recently, RL-based policy has emerged as a promising alternative due to its ability to directly map observations to actions, reducing the need for detailed system knowledge and actuation constraints. However, a significant challenge remains in bridging the sim-to-real gap, where RL-based policies often experience instability when deployed in real world. In this paper, we investigate key factors for learning robust RL-based control policies that are capable of zero-shot deployment in real-world quadrotors. We identify five critical factors and we develop a PPO-based training framework named SimpleFlight, which integrates these five techniques. We validate the efficacy of SimpleFlight on Crazyflie quadrotor, demonstrating that it achieves more than a 50% reduction in trajectory tracking error compared to state-of-the-art RL baselines. The policy derived by SimpleFlight consistently excels across both smooth polynominal trajectories and challenging infeasible zigzag trajectories on small thrust-to-weight quadrotors. In contrast, baseline methods struggle with high-speed or infeasible trajectories. To support further research and reproducibility, we integrate SimpleFlight into a GPU-based simulator Omnidrones and provide open-source access to the code and model checkpoints. We hope SimpleFlight will offer valuable insights for advancing RL-based quadrotor control. For more details, visit our project website at https://sites.google.com/view/simpleflight/.
We compare full-space, reduced-space, and gray-box formulations for representing trained neural networks in nonlinear constrained optimization problems. We test these formulations on a transient stability-constrained, security-constrained alternating current optimal power flow (SCOPF) problem where the transient stability criteria are represented by a trained neural network surrogate. Optimization problems are implemented in JuMP and trained neural networks are embedded using a new Julia package: MathOptAI.jl. To study the bottlenecks of the three formulations, we use neural networks with up to 590 million trained parameters. The full-space formulation is bottlenecked by the linear solver used by the optimization algorithm, while the reduced-space formulation is bottlenecked by the algebraic modeling environment and derivative computations. The gray-box formulation is the most scalable and is capable of solving with the largest neural networks tested. It is bottlenecked by evaluation of the neural network's outputs and their derivatives, which may be accelerated with a graphics processing unit (GPU). Leveraging the gray-box formulation and GPU acceleration, we solve our test problem with our largest neural network surrogate in 2.5$\times$ the time required for a simpler SCOPF problem without the stability constraint.