Reliable uncertainty estimation plays a crucial role in various safety-critical applications such as medical diagnosis and autonomous driving. In recent years, Bayesian neural networks (BayesNNs) have gained substantial research and industrial interests due to their capability to make accurate predictions with reliable uncertainty estimation. However, the algorithmic complexity and the resulting hardware performance of BayesNNs hinder their adoption in real-life applications. To bridge this gap, this paper proposes an algorithm and hardware co-design framework that can generate field-programmable gate array (FPGA)-based accelerators for efficient BayesNNs. At the algorithm level, we propose novel multi-exit dropout-based BayesNNs with reduced computational and memory overheads while achieving high accuracy and quality of uncertainty estimation. At the hardware level, this paper introduces a transformation framework that can generate FPGA-based accelerators for the proposed efficient multi-exit BayesNNs. Several optimization techniques such as the mix of spatial and temporal mappings are introduced to reduce resource consumption and improve the overall hardware performance. Comprehensive experiments demonstrate that our approach can achieve higher energy efficiency compared to CPU, GPU, and other state-of-the-art hardware implementations. To support the future development of this research, we have open-sourced our code at: https://github.com/os-hxfan/MCME_FPGA_Acc.git
DC Optimal Power Flow (DCOPF) is a key operational tool for power system operators, and it is embedded as a subproblem in many challenging optimization problems (e.g., line switching). However, traditional CPU-based solve routines (e.g., simplex) have saturated in speed and are hard to parallelize. This paper focuses on solving DCOPF problems using gradient-based routines on Graphics Processing Units (GPUs), which have massive parallelization capability. To formulate these problems, we pose a Lagrange dual associated with DCOPF (linear and quadratic cost curves), and then we explicitly solve the inner (primal) minimization problem with a dual norm. The resulting dual problem can be efficiently iterated using projected gradient ascent. After solving the dual problem on both CPUs and GPUs to find tight lower bounds, we benchmark against Gurobi and MOSEK, comparing convergence speed and tightness on the IEEE 2000, 4601, and 10000 bus systems. We provide reliable and tight lower bounds for these problems with, at best, 5.4x speedup over a conventional solver.
Low Rank Adaptation (LoRA) has gained massive attention in the recent generative AI research. One of the main advantages of LoRA is its ability to be fused with pretrained models, adding no overhead during inference. However, from a mobile deployment standpoint, we can either avoid inference overhead in the fused mode but lose the ability to switch adapters rapidly, or suffer significant (up to 30% higher) inference latency while enabling rapid switching in the unfused mode. LoRA also exhibits concept-loss when multiple adapters are used concurrently. In this paper, we propose Sparse High Rank Adapters (SHiRA), a new paradigm which incurs no inference overhead, enables rapid switching, and significantly reduces concept-loss. Specifically, SHiRA can be trained by directly tuning only 1-2% of the base model weights while leaving others unchanged. This results in a highly sparse adapter which can be switched directly in the fused mode. We further provide theoretical and empirical insights on how high sparsity in SHiRA can aid multi-adapter fusion by reducing concept loss. Our extensive experiments on LVMs and LLMs demonstrate that finetuning only a small fraction of the parameters in the base model significantly outperforms LoRA while enabling both rapid switching and multi-adapter fusion. Finally, we provide a latency- and memory-efficient SHiRA implementation based on Parameter-Efficient Finetuning (PEFT) Library which trains at nearly the same speed as LoRA while consuming up to 16% lower peak GPU memory, thus making SHiRA easy to adopt for practical use cases. To demonstrate rapid switching benefits during inference, we show that loading SHiRA on a base model can be 5x-16x faster than LoRA fusion on a CPU.
Deep Learning Recommendation Models (DLRMs) have gained popularity in recommendation systems due to their effectiveness in handling large-scale recommendation tasks. The embedding layers of DLRMs have become the performance bottleneck due to their intensive needs on memory capacity and memory bandwidth. In this paper, we propose UpDLRM, which utilizes real-world processingin-memory (PIM) hardware, UPMEM DPU, to boost the memory bandwidth and reduce recommendation latency. The parallel nature of the DPU memory can provide high aggregated bandwidth for the large number of irregular memory accesses in embedding lookups, thus offering great potential to reduce the inference latency. To fully utilize the DPU memory bandwidth, we further studied the embedding table partitioning problem to achieve good workload-balance and efficient data caching. Evaluations using real-world datasets show that, UpDLRM achieves much lower inference time for DLRM compared to both CPU-only and CPU-GPU hybrid counterparts.
Multispectral imaging aims at recording images in different spectral bands. This is extremely beneficial in diverse discrimination applications, for example in agriculture, recycling or healthcare. One approach for snapshot multispectral imaging, which is capable of recording multispectral videos, is by using camera arrays, where each camera records a different spectral band. Since the cameras are at different spatial positions, a registration procedure is necessary to map every camera to the same view. In this paper, we present a multispectral snapshot image registration with three novel components. First, a cross spectral disparity estimation network is introduced, which is trained on a popular stereo database using pseudo spectral data augmentation. Subsequently, this disparity estimation is used to accurately detect occlusions by warping the disparity map in a layer-wise manner. Finally, these detected occlusions are reconstructed by a learned deep guided neural network, which leverages the structure from other spectral components. It is shown that each element of this registration process as well as the final result is superior to the current state of the art. In terms of PSNR, our registration achieves an improvement of over 3 dB. At the same time, the runtime is decreased by a factor of over 3 on a CPU. Additionally, the registration is executable on a GPU, where the runtime can be decreased by a factor of 111. The source code and the data is available at https://github.com/FAU-LMS/MSIR.
The increasing size of large language models (LLMs) challenges their usage on resource-constrained platforms. For example, memory on modern GPUs is insufficient to hold LLMs that are hundreds of Gigabytes in size. Offloading is a popular method to escape this constraint by storing weights of an LLM model to host CPU memory and SSD, then loading each weight to GPU before every use. In our case study of offloaded inference, we found that due to the low bandwidth between storage devices and GPU, the latency of transferring large model weights from its offloaded location to GPU memory becomes the critical bottleneck with actual compute taking nearly 0% of runtime. To effectively reduce the weight transfer latency, we propose a novel sparse format that compresses the unstructured sparse pattern of pruned LLM weights to non-zero values with high compression ratio and low decompression overhead. Endor achieves this by expressing the positions of non-zero elements with a bitmap. Compared to offloaded inference using the popular Huggingface Accelerate, applying Endor accelerates OPT-66B by 1.70x and Llama2-70B by 1.78x. When direct weight transfer from SSD to GPU is leveraged, Endor achieves 2.25x speedup on OPT-66B and 2.37x speedup on Llama2-70B.
Recently developed large language models (LLMs) such as ChatGPT, Claude, and Llama have demonstrated impressive abilities, and even surpass human-level performance in several tasks. Despite their success, the resource-intensive demands of these models, requiring significant computational power for both training and inference, limit their deployment to high-performance servers. Additionally, the extensive calculation requirements of the models often lead to increased latency in response times. With the increasing need for LLMs to operate efficiently on CPUs, research about lightweight models that are optimized for CPU inference has emerged. In this work, we introduce GEB-1.3B, a lightweight LLM trained on 550 billion tokens in both Chinese and English languages. We employ novel training techniques, including ROPE, Group-Query-Attention, and FlashAttention-2, to accelerate training while maintaining model performance. Additionally, we fine-tune the model using 10 million samples of instruction data to enhance alignment. GEB-1.3B exhibits outstanding performance on general benchmarks such as MMLU, C-Eval, and CMMLU, outperforming comparative models such as MindLLM-1.3B and TinyLLaMA-1.1B. Notably, the FP32 version of GEB-1.3B achieves commendable inference times on CPUs, with ongoing efforts to further enhance speed through advanced quantization techniques. The release of GEB-1.3B as an open-source model marks a significant contribution to the development of lightweight LLMs, promising to foster further research and innovation in the field.
Combinatorial optimization (CO) is one of the most fundamental mathematical models in real-world applications. Traditional CO solvers, such as Branch-and-Bound (B&B) solvers, heavily rely on expert-designed heuristics, which are reliable but require substantial manual tuning. Recent studies have leveraged deep learning (DL) models as an alternative to capture rich feature patterns for improved performance on GPU machines. Nonetheless, the drawbacks of high training and inference costs, as well as limited interpretability, severely hinder the adoption of DL methods in real-world applications. To address these challenges, we propose a novel deep symbolic optimization learning framework that combines their advantages. Specifically, we focus on the node selection module within B&B solvers -- namely, deep symbolic optimization for node selection (Dso4NS). With data-driven approaches, Dso4NS guides the search for mathematical expressions within the high-dimensional discrete symbolic space and then incorporates the highest-performing mathematical expressions into a solver. The data-driven model captures the rich feature information in the input data and generates symbolic expressions, while the expressions deployed in solvers enable fast inference with high interpretability. Experiments demonstrate the effectiveness of Dso4NS in learning high-quality expressions, outperforming existing approaches on a CPU machine. Encouragingly, the learned CPU-based policies consistently achieve performance comparable to state-of-the-art GPU-based approaches.
Visual sensors are not only becoming better at capturing high-quality images but also they have steadily increased their capabilities in processing data on their own on-chip. Yet the majority of VO pipelines rely on the transmission and processing of full images in a centralized unit (e.g. CPU or GPU), which often contain much redundant and low-quality information for the task. In this paper, we address the task of frame-to-frame rotational estimation but, instead of reasoning about relative motion between frames using the full images, distribute the estimation at pixel-level. In this paradigm, each pixel produces an estimate of the global motion by only relying on local information and local message-passing with neighbouring pixels. The resulting per-pixel estimates can then be communicated to downstream tasks, yielding higher-level, informative cues instead of the original raw pixel-readings. We evaluate the proposed approach on real public datasets, where we offer detailed insights about this novel technique and open-source our implementation for the future benefit of the community.
Fine-tuning large language models (LLMs) requires significant memory, often exceeding the capacity of a single GPU. A common solution to this memory challenge is offloading compute and data from the GPU to the CPU. However, this approach is hampered by the limited bandwidth of commodity hardware, which constrains communication between the CPU and GPU, and by slower matrix multiplications on the CPU. In this paper, we present an offloading framework, LSP-Offload, that enables near-native speed LLM fine-tuning on commodity hardware through learned sparse projectors. Our data-driven approach involves learning efficient sparse compressors that minimize communication with minimal precision loss. Additionally, we introduce a novel layer-wise communication schedule to maximize parallelism between communication and computation. As a result, our framework can fine-tune a 1.3 billion parameter model on a 4GB laptop GPU and a 6.7 billion parameter model on a 24GB NVIDIA RTX 4090 GPU. Compared to state-of-the-art offloading frameworks, our approach reduces end-to-end fine-tuning time by 33.1%-62.5% when converging to the same accuracy. We open source our framework at https://github.com/gulang2019/LSP-Offload.
It is extremely memory-hungry to train Large Language Models (LLM). To solve this problem, existing work exploits the combination of CPU and GPU for the training process, such as ZeRO-Offload. Such a technique largely democratizes billion-scale model training, making it possible to train with few consumer graphics cards. However, based on our observation, existing frameworks often provide coarse-grained memory management and require experienced experts in configuration tuning, leading to suboptimal hardware utilization and performance. This paper proposes ProTrain, a novel training system that intelligently balances memory usage and performance by coordinating memory, computation, and IO. ProTrain achieves adaptive memory management through Chunk-Based Model State Management and Block-Wise Activation Management, guided by a Memory-Aware Runtime Profiler without user intervention. ProTrain does not change the training algorithm and thus does not compromise accuracy. Experiments show that ProTrain improves training throughput by 1.43$\times$ to 2.71$\times$ compared to the SOTA training systems.
Large language models (LLMs) on smartphones enable real-time AI assistance and privacy-preserving, offline operation. However, resource constraints of smartphones limit current deployments to small language models (SLMs), significantly compromising their capabilities. This paper introduces PowerInfer-2, a smartphone-based framework that enables fast inference for LLMs exceeding the memory capacity. The key insight is decomposing matrix operations into neuron clusters as the basic processing unit, which enables flexible scheduling and efficient I/O-computation pipelining. PowerInfer-2 leverages this neuron-cluster-based design in both computation and storage. For computation, neuron clusters with dense activations are processed on NPU, while sparse clusters use CPU. The storage engine provides a fine-grained pipeline mechanism that coordinates cluster-level computation and I/O operations, enhanced by a segmented neuron cache to reduce I/O activities. PowerInfer-2 achieves up to a 27.8x speed increase compared to state-of-the-art frameworks. PowerInfer-2 is the first system to serve a 47B LLM on a smartphone, achieving 11.68 tokens/s. Notably, these performance improvements preserve model quality with negligible accuracy degradation.
Large-scale datacenters often experience memory failures, where Uncorrectable Errors (UEs) highlight critical malfunction in Dual Inline Memory Modules (DIMMs). Existing approaches primarily utilize Correctable Errors (CEs) to predict UEs, yet they typically neglect how these errors vary between different CPU architectures, especially in terms of Error Correction Code (ECC) applicability. In this paper, we investigate the correlation between CEs and UEs across different CPU architectures, including X86 and ARM. Our analysis identifies unique patterns of memory failure associated with each processor platform. Leveraging Machine Learning (ML) techniques on production datasets, we conduct the memory failure prediction in different processors' platforms, achieving up to 15% improvements in F1-score compared to the existing algorithm. Finally, an MLOps (Machine Learning Operations) framework is provided to consistently improve the failure prediction in the production environment.
Machine learning ensembles combine multiple base models to produce a more accurate output. They can be applied to a range of machine learning problems, including anomaly detection. In this paper, we investigate how to maximize the composability and scalability of an FPGA-based streaming ensemble anomaly detector (fSEAD). To achieve this, we propose a flexible computing architecture consisting of multiple partially reconfigurable regions, pblocks, which each implement anomaly detectors. Our proof-of-concept design supports three state-of-the-art anomaly detection algorithms: Loda, RS-Hash and xStream. Each algorithm is scalable, meaning multiple instances can be placed within a pblock to improve performance. Moreover, fSEAD is implemented using High-level synthesis (HLS), meaning further custom anomaly detectors can be supported. Pblocks are interconnected via an AXI-switch, enabling them to be composed in an arbitrary fashion before combining and merging results at run-time to create an ensemble that maximizes the use of FPGA resources and accuracy. Through utilizing reconfigurable Dynamic Function eXchange (DFX), the detector can be modified at run-time to adapt to changing environmental conditions. We compare fSEAD to an equivalent central processing unit (CPU) implementation using four standard datasets, with speed-ups ranging from $3\times$ to $8\times$.
Parameter-Efficient Fine-tuning (PEFT) facilitates the fine-tuning of Large Language Models (LLMs) under limited resources. However, the fine-tuning performance with PEFT on complex, knowledge-intensive tasks is limited due to the constrained model capacity, which originates from the limited number of additional trainable parameters. To overcome this limitation, we introduce a novel mechanism that fine-tunes LLMs with adapters of larger size yet memory-efficient. This is achieved by leveraging the inherent activation sparsity in the Feed-Forward Networks (FFNs) of LLMs and utilizing the larger capacity of Central Processing Unit (CPU) memory compared to Graphics Processing Unit (GPU). We store and update the parameters of larger adapters on the CPU. Moreover, we employ a Mixture of Experts (MoE)-like architecture to mitigate unnecessary CPU computations and reduce the communication volume between the GPU and CPU. This is particularly beneficial over the limited bandwidth of PCI Express (PCIe). Our method can achieve fine-tuning results comparable to those obtained with larger memory capacities, even when operating under more limited resources such as a 24GB memory single GPU setup, with acceptable loss in training efficiency. Our codes are available at https://github.com/CURRENTF/MEFT.
In recent times, there has been a growing interest in utilizing personalized large models on low-spec devices, such as mobile and CPU-only devices. However, utilizing a personalized large model in the on-device is inefficient, and sometimes limited due to computational cost. To tackle the problem, this paper presents the weights separation method to minimize on-device model weights using parameter-efficient fine-tuning methods. Moreover, some people speak multiple languages in an utterance, as known as code-switching, the personalized ASR model is necessary to address such cases. However, current multilingual speech recognition models are limited to recognizing a single language within each utterance. To tackle this problem, we propose code-switching speech recognition models that incorporate fine-tuned monolingual and multilingual speech recognition models. Additionally, we introduce a gated low-rank adaptation(GLoRA) for parameter-efficient fine-tuning with minimal performance degradation. Our experiments, conducted on Korean-English code-switching datasets, demonstrate that fine-tuning speech recognition models for code-switching surpasses the performance of traditional code-switching speech recognition models trained from scratch. Furthermore, GLoRA enhances parameter-efficient fine-tuning performance compared to conventional LoRA.
The majority of existing speech bandwidth extension (BWE) methods operate under the constraint of fixed source and target sampling rates, which limits their flexibility in practical applications. In this paper, we propose a multi-stage speech BWE model named MS-BWE, which can handle a set of source and target sampling rate pairs and achieve flexible extensions of frequency bandwidth. The proposed MS-BWE model comprises a cascade of BWE blocks, with each block featuring a dual-stream architecture to realize amplitude and phase extension, progressively painting the speech frequency bands stage by stage. The teacher-forcing strategy is employed to mitigate the discrepancy between training and inference. Experimental results demonstrate that our proposed MS-BWE is comparable to state-of-the-art speech BWE methods in speech quality. Regarding generation efficiency, the one-stage generation of MS-BWE can achieve over one thousand times real-time on GPU and about sixty times on CPU.
The verification throughput is becoming a major challenge bottleneck, since the complexity and size of SoC designs are still ever increasing. Simply adding more CPU cores and running more tests in parallel will not scale anymore. This paper discusses various methods of improving verification throughput: ranking and the new machine learning (ML) based technology introduced by Cadence i.e. Xcelium ML. Both methods aim at getting comparable coverage in less CPU time by applying more efficient stimulus. Ranking selects specific seeds that simply turned out to come up with the largest coverage in previous simulations, while Xcelium ML generates optimized patterns as a result of finding correlations between randomization points and achieved coverage of previous regressions. Quantified results as well as pros & cons of each approach are discussed in this paper at the example of three actual industry projects. Both Xcelium ML and Ranking methods gave comparable compression & speedup factors around 3 consistently. But the optimized ML based regressions simulated new random scenarios occasionally producing a coverage regain of more than 100%. Finally, a methodology is proposed to use Xcelium ML efficiently throughout the product development.
High-performance Transformer trackers have shown excellent results, yet they often bear a heavy computational load. Observing that a smaller input can immediately and conveniently reduce computations without changing the model, an easy solution is to adopt the low-resolution input for efficient Transformer tracking. Albeit faster, this hurts tracking accuracy much due to information loss in low resolution tracking. In this paper, we aim to mitigate such information loss to boost the performance of the low-resolution Transformer tracking via dual knowledge distillation from a frozen high-resolution (but not a larger) Transformer tracker. The core lies in two simple yet effective distillation modules, comprising query-key-value knowledge distillation (QKV-KD) and discrimination knowledge distillation (Disc-KD), across resolutions. The former, from the global view, allows the low-resolution tracker to inherit the features and interactions from the high-resolution tracker, while the later, from the target-aware view, enhances the target-background distinguishing capacity via imitating discriminative regions from its high-resolution counterpart. With the dual knowledge distillation, our Low-Resolution Transformer Tracker (LoReTrack) enjoys not only high efficiency owing to reduced computation but also enhanced accuracy by distilling knowledge from the high-resolution tracker. In extensive experiments, LoReTrack with a 256x256 resolution consistently improves baseline with the same resolution, and shows competitive or even better results compared to 384x384 high-resolution Transformer tracker, while running 52% faster and saving 56% MACs. Moreover, LoReTrack is resolution-scalable. With a 128x128 resolution, it runs 25 fps on a CPU with 64.9%/46.4% SUC scores on LaSOT/LaSOText, surpassing all other CPU real-time trackers. Code will be released.
Background. The recent surge in phishing attacks keeps undermining the effectiveness of the traditional anti-phishing blacklist approaches. On-device anti-phishing solutions are gaining popularity as they offer faster phishing detection locally. Aim. We aim to eliminate the delay in recognizing and recording phishing campaigns in databases via on-device solutions that identify phishing sites immediately when encountered by the user rather than waiting for a web crawler's scan to finish. Additionally, utilizing operating system-specific resources and frameworks, we aim to minimize the impact on system performance and depend on local processing to protect user privacy. Method. We propose a phishing detection solution that uses a combination of computer vision and on-device machine learning models to analyze websites in real time. Our reference-based approach analyzes the visual content of webpages, identifying phishing attempts through layout analysis, credential input areas detection, and brand impersonation criteria combination. Results. Our case study shows it's feasible to perform background processing on-device continuously, for the case of the web browser requiring the resource use of 16% of a single CPU core and less than 84MB of RAM on Apple M1 while maintaining the accuracy of brand logo detection at 46.6% (comparable with baselines), and of Credential Requiring Page detection at 98.1% (improving the baseline by 3.1%), within the test dataset. Conclusions. Our results demonstrate the potential of on-device, real-time phishing detection systems to enhance cybersecurity defensive technologies and extend the scope of phishing detection to more similar regions of interest, e.g., email clients and messenger windows.