Simulation speed matters for neuroscientific research: this includes not only how quickly the simulated model time of a large-scale spiking neuronal network progresses, but also how long it takes to instantiate the network model in computer memory. On the hardware side, acceleration via highly parallel GPUs is being increasingly utilized. On the software side, code generation approaches ensure highly optimized code, at the expense of repeated code regeneration and recompilation after modifications to the network model. Aiming for a greater flexibility with respect to iterative model changes, here we propose a new method for creating network connections interactively, dynamically, and directly in GPU memory through a set of commonly used high-level connection rules. We validate the simulation performance with both consumer and data center GPUs on two neuroscientifically relevant models: a cortical microcircuit of about 77,000 leaky-integrate-and-fire neuron models and 300 million static synapses, and a two-population network recurrently connected using a variety of connection rules. With our proposed ad hoc network instantiation, both network construction and simulation times are comparable or shorter than those obtained with other state-of-the-art simulation technologies, while still meeting the flexibility demands of explorative network modeling.
This paper presents measures to reduce the computation time of automotive synthetic aperture radar (SAR) imaging to achieve real-time capability. For this, the image formation, which is based on the Back-Projection algorithm, was thoroughly analyzed. Various optimizations were individually tested and analyzed on graphics processing units (GPU). Apart from the time reduction gained from these measures, the data size needed for processing was also drastically decreased. With a combination of all measures, a high-resolution SAR image of 30 m by 30 m that combines 8192 chirps can be reconstructed in less than 30 ms using a standard GPU. It is thus demonstrated that a real-time implementation of automotive SAR is possible.
Large Language Models (LLMs) have revolutionized Natural Language Processing (NLP) but demand massive GPU resources for training. Lowering the threshold for LLMs training would encourage greater participation from researchers, benefiting both academia and society. While existing approaches have focused on parameter-efficient fine-tuning, which tunes or adds a small number of parameters, few have addressed the challenge of tuning the full parameters of LLMs with limited resources. In this work, we propose a new optimizer, LOw-Memory Optimization (LOMO), which fuses the gradient computation and the parameter update in one step to reduce memory usage. By integrating LOMO with existing memory saving techniques, we reduce memory usage to 10.8% compared to the standard approach (DeepSpeed solution). Consequently, our approach enables the full parameter fine-tuning of a 65B model on a single machine with 8 RTX 3090, each with 24GB memory.Code and data are available at https://github.com/OpenLMLab/LOMO.
Color re-indexing is one of main approaches for improving the loss-less compression of color indexed images. Zero-order entropy reduction of indexes matrix is the key to obtain high compression ratio. However, obtaining the optimal re-indexed palette is a challenging problem that cannot be solved by brute-force approaches. In this paper we propose a novel re-indexing approach where the Travelling Salesman Problem is solved through Ant Colony Optimization. Our method is proved to achieve high quality results by outperforming state-of-art ones in term of compression gain. Additionally, we exploit clustering and GPU computing to make our solution extremely fast.
This paper introduces the Fair Fairness Benchmark (\textsf{FFB}), a benchmarking framework for in-processing group fairness methods. Ensuring fairness in machine learning is important for ethical compliance. However, there exist challenges in comparing and developing fairness methods due to inconsistencies in experimental settings, lack of accessible algorithmic implementations, and limited extensibility of current fairness packages and tools. To address these issues, we introduce an open-source standardized benchmark for evaluating in-processing group fairness methods and provide a comprehensive analysis of state-of-the-art methods to ensure different notions of group fairness. This work offers the following key contributions: the provision of flexible, extensible, minimalistic, and research-oriented open-source code; the establishment of unified fairness method benchmarking pipelines; and extensive benchmarking, which yields key insights from $\mathbf{45,079}$ experiments, $\mathbf{14,428}$ GPU hours. We believe that our work will significantly facilitate the growth and development of the fairness research community.
We introduce a novel superpoint-based transformer architecture for efficient semantic segmentation of large-scale 3D scenes. Our method incorporates a fast algorithm to partition point clouds into a hierarchical superpoint structure, which makes our preprocessing 7 times faster than existing superpoint-based approaches. Additionally, we leverage a self-attention mechanism to capture the relationships between superpoints at multiple scales, leading to state-of-the-art performance on three challenging benchmark datasets: S3DIS (76.0% mIoU 6-fold validation), KITTI-360 (63.5% on Val), and DALES (79.6%). With only 212k parameters, our approach is up to 200 times more compact than other state-of-the-art models while maintaining similar performance. Furthermore, our model can be trained on a single GPU in 3 hours for a fold of the S3DIS dataset, which is 7x to 70x fewer GPU-hours than the best-performing methods. Our code and models are accessible at github.com/drprojects/superpoint_transformer.
This paper proposes the transition-net, a robust transition strategy that expands the versatility of robot locomotion in the real-world setting. To this end, we start by distributing the complexity of different gaits into dedicated locomotion policies applicable to real-world robots. Next, we expand the versatility of the robot by unifying the policies with robust transitions into a single coherent meta-controller by examining the latent state representations. Our approach enables the robot to iteratively expand its skill repertoire and robustly transition between any policy pair in a library. In our framework, adding new skills does not introduce any process that alters the previously learned skills. Moreover, training of a locomotion policy takes less than an hour with a single consumer GPU. Our approach is effective in the real-world and achieves a 19% higher average success rate for the most challenging transition pairs in our experiments compared to existing approaches.
To accelerate distributed training, many gradient compression methods have been proposed to alleviate the communication bottleneck in synchronous stochastic gradient descent (S-SGD), but their efficacy in real-world applications still remains unclear. In this work, we first evaluate the efficiency of three representative compression methods (quantization with Sign-SGD, sparsification with Top-k SGD, and low-rank with Power-SGD) on a 32-GPU cluster. The results show that they cannot always outperform well-optimized S-SGD or even worse due to their incompatibility with three key system optimization techniques (all-reduce, pipelining, and tensor fusion) in S-SGD. To this end, we propose a novel gradient compression method, called alternate compressed Power-SGD (ACP-SGD), which alternately compresses and communicates low-rank matrices. ACP-SGD not only significantly reduces the communication volume, but also enjoys the three system optimizations like S-SGD. Compared with Power-SGD, the optimized ACP-SGD can largely reduce the compression and communication overheads, while achieving similar model accuracy. In our experiments, ACP-SGD achieves an average of 4.06x and 1.43x speedups over S-SGD and Power-SGD, respectively, and it consistently outperforms other baselines across different setups (from 8 GPUs to 64 GPUs and from 1Gb/s Ethernet to 100Gb/s InfiniBand).
We present a novel model for Tracking Any Point (TAP) that effectively tracks any queried point on any physical surface throughout a video sequence. Our approach employs two stages: (1) a matching stage, which independently locates a suitable candidate point match for the query point on every other frame, and (2) a refinement stage, which updates both the trajectory and query features based on local correlations. The resulting model surpasses all baseline methods by a significant margin on the TAP-Vid benchmark, as demonstrated by an approximate 20% absolute average Jaccard (AJ) improvement on DAVIS. Our model facilitates fast inference on long and high-resolution video sequences. On a modern GPU, our implementation has the capacity to track points faster than real-time, and can be flexibly extended to higher-resolution videos. Given the high-quality trajectories extracted from a large dataset, we demonstrate a proof-of-concept diffusion model which generates trajectories from static images, enabling plausible animations. Visualizations, source code, and pretrained models can be found on our project webpage.
Finding optimal channel dimensions (i.e., the number of filters in DNN layers) is essential to design DNNs that perform well under computational resource constraints. Recent work in neural architecture search aims at automating the optimization of the DNN model implementation. However, existing neural architecture search methods for channel dimensions rely on fixed search spaces, which prevents achieving an efficient and fully automated solution. In this work, we propose a novel differentiable neural architecture search method with an efficient dynamic channel allocation algorithm to enable a flexible search space for channel dimensions. We show that the proposed framework is able to find DNN architectures that are equivalent to previous methods in task accuracy and inference latency for the CIFAR-10 dataset with an improvement of $1.3-1.7\times$ in GPU-hours and $1.5-1.7\times$ in the memory requirements during the architecture search stage. Moreover, the proposed frameworks do not require a well-engineered search space a priori, which is an important step towards fully automated design of DNN architectures.
This paper presents a comprehensive benchmarking suite tailored to offline safe reinforcement learning (RL) challenges, aiming to foster progress in the development and evaluation of safe learning algorithms in both the training and deployment phases. Our benchmark suite contains three packages: 1) expertly crafted safe policies, 2) D4RL-styled datasets along with environment wrappers, and 3) high-quality offline safe RL baseline implementations. We feature a methodical data collection pipeline powered by advanced safe RL algorithms, which facilitates the generation of diverse datasets across 38 popular safe RL tasks, from robot control to autonomous driving. We further introduce an array of data post-processing filters, capable of modifying each dataset's diversity, thereby simulating various data collection conditions. Additionally, we provide elegant and extensible implementations of prevalent offline safe RL algorithms to accelerate research in this area. Through extensive experiments with over 50000 CPU and 800 GPU hours of computations, we evaluate and compare the performance of these baseline algorithms on the collected datasets, offering insights into their strengths, limitations, and potential areas of improvement. Our benchmarking framework serves as a valuable resource for researchers and practitioners, facilitating the development of more robust and reliable offline safe RL solutions in safety-critical applications. The benchmark website is available at \url{www.offline-saferl.org}.
Sparse data are common. The traditional ``handcrafted'' features are often sparse. Embedding vectors from trained models can also be very sparse, for example, embeddings trained via the ``ReLu'' activation function. In this paper, we report our exploration of efficient search in sparse data with graph-based ANN algorithms (e.g., HNSW, or SONG which is the GPU version of HNSW), which are popular in industrial practice, e.g., search and ads (advertising). We experiment with the proprietary ads targeting application, as well as benchmark public datasets. For ads targeting, we train embeddings with the standard ``cosine two-tower'' model and we also develop the ``chi-square two-tower'' model. Both models produce (highly) sparse embeddings when they are integrated with the ``ReLu'' activation function. In EBR (embedding-based retrieval) applications, after we the embeddings are trained, the next crucial task is the approximate near neighbor (ANN) search for serving. While there are many ANN algorithms we can choose from, in this study, we focus on the graph-based ANN algorithm (e.g., HNSW-type). Sparse embeddings should help improve the efficiency of EBR. One benefit is the reduced memory cost for the embeddings. The other obvious benefit is the reduced computational time for evaluating similarities, because, for graph-based ANN algorithms such as HNSW, computing similarities is often the dominating cost. In addition to the effort on leveraging data sparsity for storage and computation, we also integrate ``sign cauchy random projections'' (SignCRP) to hash vectors to bits, to further reduce the memory cost and speed up the ANN search. In NIPS'13, SignCRP was proposed to hash the chi-square similarity, which is a well-adopted nonlinear kernel in NLP and computer vision. Therefore, the chi-square two-tower model, SignCRP, and HNSW are now tightly integrated.
Generative Large Language Models (LLMs) have demonstrated remarkable results for a wide range of tasks. However, deploying these models for inference has been a significant challenge due to their unprecedented resource requirements. This has forced existing deployment frameworks to use multi-GPU inference pipelines, which are often complex and costly, or to use smaller and less performant models. In this work, we demonstrate that the main bottleneck for generative inference with LLMs is memory bandwidth, rather than compute, specifically for single batch inference. While quantization has emerged as a promising solution by representing weights with reduced precision, previous efforts have often resulted in notable performance degradation. To address this, we introduce SqueezeLLM, a post-training quantization framework that not only enables lossless compression to ultra-low precisions of up to 3-bit, but also achieves higher quantization performance under the same memory constraint. Our framework incorporates two novel ideas: (i) sensitivity-based non-uniform quantization, which searches for the optimal bit precision assignment based on second-order information; and (ii) the Dense-and-Sparse decomposition that stores outliers and sensitive weight values in an efficient sparse format. When applied to the LLaMA models, our 3-bit quantization significantly reduces the perplexity gap from the FP16 baseline by up to 2.1x as compared to the state-of-the-art methods with the same memory requirement. Furthermore, when deployed on an A6000 GPU, our quantized models achieve up to 2.3x speedup compared to the baseline. Our code is available at https://github.com/SqueezeAILab/SqueezeLLM.
Poisson's equation plays an important role in modeling many physical systems. In electrostatic self-consistent low-temperature plasma (LTP) simulations, Poisson's equation is solved at each simulation time step, which can amount to a significant computational cost for the entire simulation. In this paper, we describe the development of a generic machine-learned Poisson solver specifically designed for the requirements of LTP simulations in complex 2D reactor geometries on structured Cartesian grids. Here, the reactor geometries can consist of inner electrodes and dielectric materials as often found in LTP simulations. The approach leverages a hybrid CNN-transformer network architecture in combination with a weighted multiterm loss function. We train the network using highly-randomized synthetic data to ensure the generalizability of the learned solver to unseen reactor geometries. The results demonstrate that the learned solver is able to produce quantitatively and qualitatively accurate solutions. Furthermore, it generalizes well on new reactor geometries such as reference geometries found in the literature. To increase the numerical accuracy of the solutions required in LTP simulations, we employ a conventional iterative solver to refine the raw predictions, especially to recover the high-frequency features not resolved by the initial prediction. With this, the proposed learned Poisson solver provides the required accuracy and is potentially faster than a pure GPU-based conventional iterative solver. This opens up new possibilities for developing a generic and high-performing learned Poisson solver for LTP systems in complex geometries.
We present Galactic, a large-scale simulation and reinforcement-learning (RL) framework for robotic mobile manipulation in indoor environments. Specifically, a Fetch robot (equipped with a mobile base, 7DoF arm, RGBD camera, egomotion, and onboard sensing) is spawned in a home environment and asked to rearrange objects - by navigating to an object, picking it up, navigating to a target location, and then placing the object at the target location. Galactic is fast. In terms of simulation speed (rendering + physics), Galactic achieves over 421,000 steps-per-second (SPS) on an 8-GPU node, which is 54x faster than Habitat 2.0 (7699 SPS). More importantly, Galactic was designed to optimize the entire rendering + physics + RL interplay since any bottleneck in the interplay slows down training. In terms of simulation+RL speed (rendering + physics + inference + learning), Galactic achieves over 108,000 SPS, which 88x faster than Habitat 2.0 (1243 SPS). These massive speed-ups not only drastically cut the wall-clock training time of existing experiments, but also unlock an unprecedented scale of new experiments. First, Galactic can train a mobile pick skill to >80% accuracy in under 16 minutes, a 100x speedup compared to the over 24 hours it takes to train the same skill in Habitat 2.0. Second, we use Galactic to perform the largest-scale experiment to date for rearrangement using 5B steps of experience in 46 hours, which is equivalent to 20 years of robot experience. This scaling results in a single neural network composed of task-agnostic components achieving 85% success in GeometricGoal rearrangement, compared to 0% success reported in Habitat 2.0 for the same approach. The code is available at github.com/facebookresearch/galactic.
We train a small message-passing graph neural network to predict Hamiltonian cycles on Erd\H{o}s-R\'enyi random graphs in a critical regime. It outperforms existing hand-crafted heuristics after about 2.5 hours of training on a single GPU. Our findings encourage an alternative approach to solving computationally demanding (NP-hard) problems arising in practice. Instead of devising a heuristic by hand, one can train it end-to-end using a neural network. This has several advantages. Firstly, it is relatively quick and requires little problem-specific knowledge. Secondly, the network can adjust to the distribution of training samples, improving the performance on the most relevant problem instances. The model is trained using supervised learning on artificially created problem instances; this training procedure does not use an existing solver to produce the supervised signal. Finally, the model generalizes well to larger graph sizes and retains reasonable performance even on graphs eight times the original size.
A neural network consisting of piecewise affine building blocks, such as fully-connected layers and ReLU activations, is itself a piecewise affine function supported on a polyhedral complex. This complex has been previously studied to characterize theoretical properties of neural networks, but, in practice, extracting it remains a challenge due to its high combinatorial complexity. A natural idea described in previous works is to subdivide the regions via intersections with hyperplanes induced by each neuron. However, we argue that this view leads to computational redundancy. Instead of regions, we propose to subdivide edges, leading to a novel method for polyhedral complex extraction. A key to this are sign-vectors, which encode the combinatorial structure of the complex. Our approach allows to use standard tensor operations on a GPU, taking seconds for millions of cells on a consumer grade machine. Motivated by the growing interest in neural shape representation, we use the speed and differentiability of our method to optimize geometric properties of the complex. The code is available at https://github.com/arturs-berzins/relu_edge_subdivision .
Optimal transport (OT) is a powerful framework to compare probability measures, a fundamental task in many statistical and machine learning problems. Substantial advances have been made in designing OT variants which are either computationally and statistically more efficient or robust. Among them, sliced OT distances have been extensively used to mitigate optimal transport's cubic algorithmic complexity and curse of dimensionality. In parallel, unbalanced OT was designed to allow comparisons of more general positive measures, while being more robust to outliers. In this paper, we bridge the gap between those two concepts and develop a general framework for efficiently comparing positive measures. We notably formulate two different versions of sliced unbalanced OT, and study the associated topology and statistical properties. We then develop a GPU-friendly Frank-Wolfe like algorithm to compute the corresponding loss functions, and show that the resulting methodology is modular as it encompasses and extends prior related work. We finally conduct an empirical analysis of our loss functions and methodology on both synthetic and real datasets, to illustrate their computational efficiency, relevance and applicability to real-world scenarios including geophysical data.
We introduce a library called Push that takes a probabilistic programming approach to Bayesian deep learning (BDL). This library enables concurrent execution of BDL inference algorithms on multi-GPU hardware for neural network (NN) models. To accomplish this, Push introduces an abstraction that represents an input NN as a particle. Push enables easy creation of particles so that an input NN can be replicated and particles can communicate asynchronously so that a variety of parameter updates can be expressed, including common BDL algorithms. Our hope is that Push lowers the barrier to experimenting with BDL by streamlining the scaling of particles across GPUs. We evaluate the scaling behavior of particles on single-node multi-GPU devices on vision and scientific machine learning (SciML) tasks.
Artificial intelligence (AI) has been widely used in bioimage image analysis nowadays, but the efficiency of AI models, like the energy consumption and latency is not ignorable due to the growing model size and complexity, as well as the fast-growing analysis needs in modern biomedical studies. Like we can compress large images for efficient storage and sharing, we can also compress the AI models for efficient applications and deployment. In this work, we present EfficientBioAI, a plug-and-play toolbox that can compress given bioimaging AI models for them to run with significantly reduced energy cost and inference time on both CPU and GPU, without compromise on accuracy. In some cases, the prediction accuracy could even increase after compression, since the compression procedure could remove redundant information in the model representation and therefore reduce over-fitting. From four different bioimage analysis applications, we observed around 2-5 times speed-up during inference and 30-80$\%$ saving in energy. Cutting the runtime of large scale bioimage analysis from days to hours or getting a two-minutes bioimaging AI model inference done in near real-time will open new doors for method development and biomedical discoveries. We hope our toolbox will facilitate resource-constrained bioimaging AI and accelerate large-scale AI-based quantitative biological studies in an eco-friendly way, as well as stimulate further research on the efficiency of bioimaging AI.