PyTorchKR Papers

최신 논문

총 2,153편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2403.20222공식 코드 · Papers with Code

Shallow Cross-Encoders for Low-Latency Retrieval

Transformer-based Cross-Encoders achieve state-of-the-art effectiveness in text retrieval. However, Cross-Encoders based on large transformer models (such as BERT or T5) are computationally expensive and allow for scoring only a small number of documents within a reasonably small latency window. However, keeping search latencies low is important for user satisfaction and energy usage. In this paper, we show that weaker shallow transformer models (i.e., transformers with a limited number of layers) actually perform better than full-scale models when constrained to these practical low-latency settings since they can estimate the relevance of more documents in the same time budget. We further show that shallow transformers may benefit from the generalized Binary Cross-Entropy (gBCE) training scheme, which has recently demonstrated success for recommendation tasks. Our experiments with TREC Deep Learning passage ranking query sets demonstrate significant improvements in shallow and full-scale models in low-latency scenarios. For example, when the latency limit is 25ms per query, MonoBERT-Large (a cross-encoder based on a full-scale BERT model) is only able to achieve NDCG@10 of 0.431 on TREC DL 2019, while TinyBERT-gBCE (a cross-encoder based on TinyBERT trained with gBCE) reaches NDCG@10 of 0.652, a +51% gain over MonoBERT-Large. We also show that shallow Cross-Encoders are effective even when used without a GPU (e.g., with CPU inference, NDCG@10 decreases only by 3% compared to GPU inference with 50ms latency), which makes Cross-Encoders practical to run even without specialized hardware acceleration.

arXiv:2403.20041

Transformer-Lite: High-efficiency Deployment of Large Language Models on Mobile Phone GPUs

The Large Language Model (LLM) is widely employed for tasks such as intelligent assistants, text summarization, translation, and multi-modality on mobile phones. However, the current methods for on-device LLM deployment maintain slow inference speed, which causes poor user experience. To facilitate high-efficiency LLM deployment on device GPUs, we propose four optimization techniques: (a) a symbolic expression-based approach to support dynamic shape model inference; (b) operator optimizations and execution priority setting to enhance inference speed and reduce phone lagging; (c) an FP4 quantization method termed M0E4 to reduce dequantization overhead; (d) a sub-tensor-based technique to eliminate the need for copying KV cache after LLM inference. Furthermore, we implement these methods in our mobile inference engine, Transformer-Lite, which is compatible with both Qualcomm and MTK processors. We evaluated Transformer-Lite's performance using LLMs with varied architectures and parameters ranging from 2B to 14B. Specifically, we achieved prefill and decoding speeds of 121 token/s and 14 token/s for ChatGLM2 6B, and 330 token/s and 30 token/s for smaller Gemma 2B, respectively. Compared with CPU-based FastLLM and GPU-based MLC-LLM, our engine attains over 10x speedup for the prefill speed and 2~3x speedup for the decoding speed.

arXiv:2403.19132

Meta-Heuristic Fronthaul Bit Allocation for Cell-free Massive MIMO Systems

Limited capacity of fronthaul links in a cell-free massive multiple-input multiple-output (MIMO) system can cause quantization errors at a central processing unit (CPU) during data transmission, complicating the centralized rate optimization problem. Addressing this challenge, we propose a harmony search (HS)-based algorithm that renders the combinatorial non-convex problem tractable. One of the distinctive features of our algorithm is its hierarchical structure: it first allocates resources at the access point (AP) level and subsequently optimizes for user equipment (UE), ensuring a more efficient and structured approach to resource allocation. Our proposed algorithm deals with rigorous conditions, such as asymmetric fronthaul bit allocation and distinct quantization error levels at each AP, which were not considered in previous works. We derive a closed-form expression of signal-to-interference-plusnoise ratio (SINR), in which additive quantization noise model (AQNM) based distortion error is taken into account, to define the mathematical expression of spectral efficiency (SE) for each UE. Also, we provide analyses on computational complexity and convergence to investigate the practicality of proposed algorithm. By leveraging various performance metrics such as total SE and max-min fairness, we demonstrate that the proposed algorithm can adaptively optimize the fronthaul bit allocation depending on system requirements. Finally, simulation results show that the proposed algorithm can achieve satisfactory performance while maintaining low computational complexity, as compared to the exhaustive search method

arXiv:2403.19421공식 코드 · Papers with Code

Scaling up ridge regression for brain encoding in a massive individual fMRI dataset

Brain encoding with neuroimaging data is an established analysis aimed at predicting human brain activity directly from complex stimuli features such as movie frames. Typically, these features are the latent space representation from an artificial neural network, and the stimuli are image, audio, or text inputs. Ridge regression is a popular prediction model for brain encoding due to its good out-of-sample generalization performance. However, training a ridge regression model can be highly time-consuming when dealing with large-scale deep functional magnetic resonance imaging (fMRI) datasets that include many space-time samples of brain activity. This paper evaluates different parallelization techniques to reduce the training time of brain encoding with ridge regression on the CNeuroMod Friends dataset, one of the largest deep fMRI resource currently available. With multi-threading, our results show that the Intel Math Kernel Library (MKL) significantly outperforms the OpenBLAS library, being 1.9 times faster using 32 threads on a single machine. We then evaluated the Dask multi-CPU implementation of ridge regression readily available in scikit-learn (MultiOutput), and we proposed a new "batch" version of Dask parallelization, motivated by a time complexity analysis. In line with our theoretical analysis, MultiOutput parallelization was found to be impractical, i.e., slower than multi-threading on a single machine. In contrast, the Batch-MultiOutput regression scaled well across compute nodes and threads, providing speed-ups of up to 33 times with 8 compute nodes and 32 threads compared to a single-threaded scikit-learn execution. Batch parallelization using Dask thus emerges as a scalable approach for brain encoding with ridge regression on high-performance computing systems using scikit-learn and large fMRI datasets.

arXiv:2403.19238공식 코드 · Papers with Code

Taming Lookup Tables for Efficient Image Retouching

The widespread use of high-definition screens in edge devices, such as end-user cameras, smartphones, and televisions, is spurring a significant demand for image enhancement. Existing enhancement models often optimize for high performance while falling short of reducing hardware inference time and power consumption, especially on edge devices with constrained computing and storage resources. To this end, we propose Image Color Enhancement Lookup Table (ICELUT) that adopts LUTs for extremely efficient edge inference, without any convolutional neural network (CNN). During training, we leverage pointwise (1x1) convolution to extract color information, alongside a split fully connected layer to incorporate global information. Both components are then seamlessly converted into LUTs for hardware-agnostic deployment. ICELUT achieves near-state-of-the-art performance and remarkably low power consumption. We observe that the pointwise network structure exhibits robust scalability, upkeeping the performance even with a heavily downsampled 32x32 input image. These enable ICELUT, the first-ever purely LUT-based image enhancer, to reach an unprecedented speed of 0.4ms on GPU and 7ms on CPU, at least one order faster than any CNN solution. Codes are available at https://github.com/Stephen0808/ICELUT.

arXiv:2403.17154공식 코드 · Papers with Code

On the Impact of Black-box Deployment Strategies for Edge AI on Latency and Model Performance

Deciding what combination of operators to use across the Edge AI tiers to achieve specific latency and model performance requirements is an open question for MLOps engineers. This study aims to empirically assess the accuracy vs inference time trade-off of different black-box Edge AI deployment strategies, i.e., combinations of deployment operators and deployment tiers. In this paper, we conduct inference experiments involving 3 deployment operators (i.e., Partitioning, Quantization, Early Exit), 3 deployment tiers (i.e., Mobile, Edge, Cloud) and their combinations on four widely used Computer-Vision models to investigate the optimal strategies from the point of view of MLOps developers. Our findings suggest that Edge deployment using the hybrid Quantization + Early Exit operator could be preferred over non-hybrid operators (Quantization/Early Exit on Edge, Partition on Mobile-Edge) when faster latency is a concern at medium accuracy loss. However, when minimizing accuracy loss is a concern, MLOps engineers should prefer using only a Quantization operator on edge at a latency reduction or increase, respectively over the Early Exit/Partition (on edge/mobile-edge) and Quantized Early Exit (on edge) operators. In scenarios constrained by Mobile CPU/RAM resources, a preference for Partitioning across mobile and edge tiers is observed over mobile deployment. For models with smaller input data samples (such as FCN), a network-constrained cloud deployment can also be a better alternative than Mobile/Edge deployment and Partitioning strategies. For models with large input data samples (ResNet, ResNext, DUC), an edge tier having higher network/computational capabilities than Cloud/Mobile can be a more viable option than Partitioning and Mobile/Cloud deployment strategies.

arXiv:2403.17312

ALISA: Accelerating Large Language Model Inference via Sparsity-Aware KV Caching

The Transformer architecture has significantly advanced natural language processing (NLP) and has been foundational in developing large language models (LLMs) such as LLaMA and OPT, which have come to dominate a broad range of NLP tasks. Despite their superior accuracy, LLMs present unique challenges in practical inference, concerning the compute and memory-intensive nature. Thanks to the autoregressive characteristic of LLM inference, KV caching for the attention layers in Transformers can effectively accelerate LLM inference by substituting quadratic-complexity computation with linear-complexity memory accesses. Yet, this approach requires increasing memory as demand grows for processing longer sequences. The overhead leads to reduced throughput due to I/O bottlenecks and even out-of-memory errors, particularly on resource-constrained systems like a single commodity GPU. In this paper, we propose ALISA, a novel algorithm-system co-design solution to address the challenges imposed by KV caching. On the algorithm level, ALISA prioritizes tokens that are most important in generating a new token via a Sparse Window Attention (SWA) algorithm. SWA introduces high sparsity in attention layers and reduces the memory footprint of KV caching at negligible accuracy loss. On the system level, ALISA employs three-phase token-level dynamical scheduling and optimizes the trade-off between caching and recomputation, thus maximizing the overall performance in resource-constrained systems. In a single GPU-CPU system, we demonstrate that under varying workloads, ALISA improves the throughput of baseline systems such as FlexGen and vLLM by up to 3X and 1.9X, respectively.

arXiv:2403.15195

FSD-Inference: Fully Serverless Distributed Inference with Scalable Cloud Communication

Serverless computing offers attractive scalability, elasticity and cost-effectiveness. However, constraints on memory, CPU and function runtime have hindered its adoption for data-intensive applications and machine learning (ML) workloads. Traditional 'server-ful' platforms enable distributed computation via fast networks and well-established inter-process communication (IPC) mechanisms such as MPI and shared memory. In the absence of such solutions in the serverless domain, parallel computation with significant IPC requirements is challenging. We present FSD-Inference, the first fully serverless and highly scalable system for distributed ML inference. We explore potential communication channels, in conjunction with Function-as-a-Service (FaaS) compute, to design a state-of-the-art solution for distributed ML within the context of serverless data-intensive computing. We introduce novel fully serverless communication schemes for ML inference workloads, leveraging both cloud-based publish-subscribe/queueing and object storage offerings. We demonstrate how publish-subscribe/queueing services can be adapted for FaaS IPC with comparable performance to object storage, while offering significantly reduced cost at high parallelism levels. We conduct in-depth experiments on benchmark DNNs of various sizes. The results show that when compared to server-based alternatives, FSD-Inference is significantly more cost-effective and scalable, and can even achieve competitive performance against optimized HPC solutions. Experiments also confirm that our serverless solution can handle large distributed workloads and leverage high degrees of FaaS parallelism.

arXiv:2403.14853공식 코드 · Papers with Code

iSpLib: A Library for Accelerating Graph Neural Networks using Auto-tuned Sparse Operations

Core computations in Graph Neural Network (GNN) training and inference are often mapped to sparse matrix operations such as sparse-dense matrix multiplication (SpMM). These sparse operations are harder to optimize by manual tuning because their performance depends significantly on the sparsity of input graphs, GNN models, and computing platforms. To address this challenge, we present iSpLib, a PyTorch-based C++ library equipped with auto-tuned sparse operations. iSpLib expedites GNN training with a cache-enabled backpropagation that stores intermediate matrices in local caches. The library offers a user-friendly Python plug-in that allows users to take advantage of our optimized PyTorch operations out-of-the-box for any existing linear algebra-based PyTorch implementation of popular GNNs (Graph Convolution Network, GraphSAGE, Graph Inference Network, etc.) with only two lines of additional code. We demonstrate that iSpLib obtains up to 27x overall training speedup compared to the equivalent PyTorch 2.1.0 and PyTorch Geometric 2.4.0 implementations on the CPU. Our library is publicly available at https://github.com/HipGraph/iSpLib (https://doi.org/10.5281/zenodo.10806511).

arXiv:2403.14290공식 코드 · Papers with Code

Exploring Green AI for Audio Deepfake Detection

The state-of-the-art audio deepfake detectors leveraging deep neural networks exhibit impressive recognition performance. Nonetheless, this advantage is accompanied by a significant carbon footprint. This is mainly due to the use of high-performance computing with accelerators and high training time. Studies show that average deep NLP model produces around 626k lbs of CO\textsubscript{2} which is equivalent to five times of average US car emission at its lifetime. This is certainly a massive threat to the environment. To tackle this challenge, this study presents a novel framework for audio deepfake detection that can be seamlessly trained using standard CPU resources. Our proposed framework utilizes off-the-shelve self-supervised learning (SSL) based models which are pre-trained and available in public repositories. In contrast to existing methods that fine-tune SSL models and employ additional deep neural networks for downstream tasks, we exploit classical machine learning algorithms such as logistic regression and shallow neural networks using the SSL embeddings extracted using the pre-trained model. Our approach shows competitive results compared to the commonly used high-carbon footprint approaches. In experiments with the ASVspoof 2019 LA dataset, we achieve a 0.90\% equal error rate (EER) with less than 1k trainable model parameters. To encourage further research in this direction and support reproducible results, the Python code will be made publicly accessible following acceptance\footnote{\href{https://github.com/sahasubhajit/Speech-Spoofing-}{GitHub link}}.

arXiv:2403.13091공식 코드 · Papers with Code

JaxUED: A simple and useable UED library in Jax

We present JaxUED, an open-source library providing minimal dependency implementations of modern Unsupervised Environment Design (UED) algorithms in Jax. JaxUED leverages hardware acceleration to obtain on the order of 100x speedups compared to prior, CPU-based implementations. Inspired by CleanRL, we provide fast, clear, understandable, and easily modifiable implementations, with the aim of accelerating research into UED. This paper describes our library and contains baseline results. Code can be found at https://github.com/DramaCow/jaxued.

arXiv:2403.13443공식 코드 · Papers with Code

Fast-Poly: A Fast Polyhedral Framework For 3D Multi-Object Tracking

3D Multi-Object Tracking (MOT) captures stable and comprehensive motion states of surrounding obstacles, essential for robotic perception. However, current 3D trackers face issues with accuracy and latency consistency. In this paper, we propose Fast-Poly, a fast and effective filter-based method for 3D MOT. Building upon our previous work Poly-MOT, Fast-Poly addresses object rotational anisotropy in 3D space, enhances local computation densification, and leverages parallelization technique, improving inference speed and precision. Fast-Poly is extensively tested on two large-scale tracking benchmarks with Python implementation. On the nuScenes dataset, Fast-Poly achieves new state-of-the-art performance with 75.8% AMOTA among all methods and can run at 34.2 FPS on a personal CPU. On the Waymo dataset, Fast-Poly exhibits competitive accuracy with 63.6% MOTA and impressive inference speed (35.5 FPS). The source code is publicly available at https://github.com/lixiaoyu2000/FastPoly.

arXiv:2403.11872

Graph Neural Network for Neutrino Physics Event Reconstruction

Liquid Argon Time Projection Chamber (LArTPC) detector technology offers a wealth of high-resolution information on particle interactions, and leveraging that information to its full potential requires sophisticated automated reconstruction techniques. This article describes NuGraph2, a Graph Neural Network (GNN) for low-level reconstruction of simulated neutrino interactions in a LArTPC detector. Simulated neutrino interactions in the MicroBooNE detector geometry are described as heterogeneous graphs, with energy depositions on each detector plane forming nodes on planar subgraphs. The network utilizes a multi-head attention message-passing mechanism to perform background filtering and semantic labelling on these graph nodes, identifying those associated with the primary physics interaction with 98.0\% efficiency and labelling them according to particle type with 94.9\% efficiency. The network operates directly on detector observables across multiple 2D representations, but utilizes a 3D-context-aware mechanism to encourage consistency between these representations. Model inference takes 0.12~s/event on a CPU, and 0.005s/event batched on a GPU. This architecture is designed to be a general-purpose solution for particle reconstruction in neutrino physics, with the potential for deployment across a broad range of detector technologies, and offers a core convolution engine that can be leveraged for a variety of tasks beyond the two described in this article.

arXiv:2403.11875

Towards Real-Time Fast Unmanned Aerial Vehicle Detection Using Dynamic Vision Sensors

Unmanned Aerial Vehicles (UAVs) are gaining popularity in civil and military applications. However, uncontrolled access to restricted areas threatens privacy and security. Thus, prevention and detection of UAVs are pivotal to guarantee confidentiality and safety. Although active scanning, mainly based on radars, is one of the most accurate technologies, it can be expensive and less versatile than passive inspections, e.g., object recognition. Dynamic vision sensors (DVS) are bio-inspired event-based vision models that leverage timestamped pixel-level brightness changes in fast-moving scenes that adapt well to low-latency object detection. This paper presents F-UAV-D (Fast Unmanned Aerial Vehicle Detector), an embedded system that enables fast-moving drone detection. In particular, we propose a setup to exploit DVS as an alternative to RGB cameras in a real-time and low-power configuration. Our approach leverages the high-dynamic range (HDR) and background suppression of DVS and, when trained with various fast-moving drones, outperforms RGB input in suboptimal ambient conditions such as low illumination and fast-moving scenes. Our results show that F-UAV-D can (i) detect drones by using less than <15 W on average and (ii) perform real-time inference (i.e., <50 ms) by leveraging the CPU and GPU nodes of our edge computer.

arXiv:2403.11651공식 코드 · Papers with Code

Overfitted image coding at reduced complexity

Overfitted image codecs offer compelling compression performance and low decoder complexity, through the overfitting of a lightweight decoder for each image. Such codecs include Cool-chic, which presents image coding performance on par with VVC while requiring around 2000 multiplications per decoded pixel. This paper proposes to decrease Cool-chic encoding and decoding complexity. The encoding complexity is reduced by shortening Cool-chic training, up to the point where no overfitting is performed at all. It is also shown that a tiny neural decoder with 300 multiplications per pixel still outperforms HEVC. A near real-time CPU implementation of this decoder is made available at https://orange-opensource.github.io/Cool-Chic/.

arXiv:2403.11472

Accelerating String-Key Learned Index Structures via Memoization-based Incremental Training

Learned indexes use machine learning models to learn the mappings between keys and their corresponding positions in key-value indexes. These indexes use the mapping information as training data. Learned indexes require frequent retrainings of their models to incorporate the changes introduced by update queries. To efficiently retrain the models, existing learned index systems often harness a linear algebraic QR factorization technique that performs matrix decomposition. This factorization approach processes all key-position pairs during each retraining, resulting in compute operations that grow linearly with the total number of keys and their lengths. Consequently, the retrainings create a severe performance bottleneck, especially for variable-length string keys, while the retrainings are crucial for maintaining high prediction accuracy and in turn, ensuring low query service latency. To address this performance problem, we develop an algorithm-hardware co-designed string-key learned index system, dubbed SIA. In designing SIA, we leverage a unique algorithmic property of the matrix decomposition-based training method. Exploiting the property, we develop a memoization-based incremental training scheme, which only requires computation over updated keys, while decomposition results of non-updated keys from previous computations can be reused. We further enhance SIA to offload a portion of this training process to an FPGA accelerator to not only relieve CPU resources for serving index queries (i.e., inference), but also accelerate the training itself. Our evaluation shows that compared to ALEX, LIPP, and SIndex, a state-of-the-art learned index systems, SIA-accelerated learned indexes offer 2.6x and 3.4x higher throughput on the two real-world benchmark suites, YCSB and Twitter cache trace, respectively.

arXiv:2403.10720

Development and Application of a Monte Carlo Tree Search Algorithm for Simulating Da Vinci Code Game Strategies

In this study, we explore the efficiency of the Monte Carlo Tree Search (MCTS), a prominent decision-making algorithm renowned for its effectiveness in complex decision environments, contingent upon the volume of simulations conducted. Notwithstanding its broad applicability, the algorithm's performance can be adversely impacted in certain scenarios, particularly within the domain of game strategy development. This research posits that the inherent branch divergence within the Da Vinci Code board game significantly impedes parallelism when executed on Graphics Processing Units (GPUs). To investigate this hypothesis, we implemented and meticulously evaluated two variants of the MCTS algorithm, specifically designed to assess the impact of branch divergence on computational performance. Our comparative analysis reveals a linear improvement in performance with the CPU-based implementation, in stark contrast to the GPU implementation, which exhibits a non-linear enhancement pattern and discernible performance troughs. These findings contribute to a deeper understanding of the MCTS algorithm's behavior in divergent branch scenarios, highlighting critical considerations for optimizing game strategy algorithms on parallel computing architectures.

arXiv:2403.09850공식 코드 · Papers with Code

MARVIS: Motion & Geometry Aware Real and Virtual Image Segmentation

Tasks such as autonomous navigation, 3D reconstruction, and object recognition near the water surfaces are crucial in marine robotics applications. However, challenges arise due to dynamic disturbances, e.g., light reflections and refraction from the random air-water interface, irregular liquid flow, and similar factors, which can lead to potential failures in perception and navigation systems. Traditional computer vision algorithms struggle to differentiate between real and virtual image regions, significantly complicating tasks. A virtual image region is an apparent representation formed by the redirection of light rays, typically through reflection or refraction, creating the illusion of an object's presence without its actual physical location. This work proposes a novel approach for segmentation on real and virtual image regions, exploiting synthetic images combined with domain-invariant information, a Motion Entropy Kernel, and Epipolar Geometric Consistency. Our segmentation network does not need to be re-trained if the domain changes. We show this by deploying the same segmentation network in two different domains: simulation and the real world. By creating realistic synthetic images that mimic the complexities of the water surface, we provide fine-grained training data for our network (MARVIS) to discern between real and virtual images effectively. By motion & geometry-aware design choices and through comprehensive experimental analysis, we achieve state-of-the-art real-virtual image segmentation performance in unseen real world domain, achieving an IoU over 78% and a F1-Score over 86% while ensuring a small computational footprint. MARVIS offers over 43 FPS (8 FPS) inference rates on a single GPU (CPU core). Our code and dataset are available here https://github.com/jiayi-wu-umd/MARVIS.

arXiv:2403.09107공식 코드 · Papers with Code

S^2MVTC: a Simple yet Efficient Scalable Multi-View Tensor Clustering

Anchor-based large-scale multi-view clustering has attracted considerable attention for its effectiveness in handling massive datasets. However, current methods mainly seek the consensus embedding feature for clustering by exploring global correlations between anchor graphs or projection matrices.In this paper, we propose a simple yet efficient scalable multi-view tensor clustering (S^2MVTC) approach, where our focus is on learning correlations of embedding features within and across views. Specifically, we first construct the embedding feature tensor by stacking the embedding features of different views into a tensor and rotating it. Additionally, we build a novel tensor low-frequency approximation (TLFA) operator, which incorporates graph similarity into embedding feature learning, efficiently achieving smooth representation of embedding features within different views. Furthermore, consensus constraints are applied to embedding features to ensure inter-view semantic consistency. Experimental results on six large-scale multi-view datasets demonstrate that S^2MVTC significantly outperforms state-of-the-art algorithms in terms of clustering performance and CPU execution time, especially when handling massive data. The code of S^2MVTC is publicly available at https://github.com/longzhen520/S2MVTC.

arXiv:2403.07921

Merino: Entropy-driven Design for Generative Language Models on IoT Devices

Generative Large Language Models (LLMs) stand as a revolutionary advancement in the modern era of artificial intelligence (AI). However, scaling down LLMs for resource-constrained hardware, such as Internet-of-Things (IoT) devices requires non-trivial efforts and domain knowledge. In this paper, we propose a novel information-entropy framework for designing mobile-friendly generative language models. The whole design procedure involves solving a mathematical programming (MP) problem, which can be done on the CPU within minutes, making it nearly zero-cost. We evaluate our designed models, termed MeRino, across fourteen NLP downstream tasks, showing their competitive performance against the state-of-the-art autoregressive transformer models under the mobile setting. Notably, MeRino achieves similar or better performance on both language modeling and zero-shot learning tasks, compared to the 350M parameter OPT while being 4.9x faster on NVIDIA Jetson Nano with 5.5x reduction in model size.