Recently, efficient Vision Transformers have shown great performance with low latency on resource-constrained devices. Conventionally, they use 4x4 patch embeddings and a 4-stage structure at the macro level, while utilizing sophisticated attention with multi-head configuration at the micro level. This paper aims to address computational redundancy at all design levels in a memory-efficient manner. We discover that using larger-stride patchify stem not only reduces memory access costs but also achieves competitive performance by leveraging token representations with reduced spatial redundancy from the early stages. Furthermore, our preliminary analyses suggest that attention layers in the early stages can be substituted with convolutions, and several attention heads in the latter stages are computationally redundant. To handle this, we introduce a single-head attention module that inherently prevents head redundancy and simultaneously boosts accuracy by parallelly combining global and local information. Building upon our solutions, we introduce SHViT, a Single-Head Vision Transformer that obtains the state-of-the-art speed-accuracy tradeoff. For example, on ImageNet-1k, our SHViT-S4 is 3.3x, 8.1x, and 2.4x faster than MobileViTv2 x1.0 on GPU, CPU, and iPhone12 mobile device, respectively, while being 1.3% more accurate. For object detection and instance segmentation on MS COCO using Mask-RCNN head, our model achieves performance comparable to FastViT-SA12 while exhibiting 3.8x and 2.0x lower backbone latency on GPU and mobile device, respectively.
Deep learning-based methods have made significant achievements in music source separation. However, obtaining good results while maintaining a low model complexity remains challenging in super wide-band music source separation. Previous works either overlook the differences in subbands or inadequately address the problem of information loss when generating subband features. In this paper, we propose SCNet, a novel frequency-domain network to explicitly split the spectrogram of the mixture into several subbands and introduce a sparsity-based encoder to model different frequency bands. We use a higher compression ratio on subbands with less information to improve the information density and focus on modeling subbands with more information. In this way, the separation performance can be significantly improved using lower computational consumption. Experiment results show that the proposed model achieves a signal to distortion ratio (SDR) of 9.0 dB on the MUSDB18-HQ dataset without using extra data, which outperforms state-of-the-art methods. Specifically, SCNet's CPU inference time is only 48% of HT Demucs, one of the previous state-of-the-art models.
This paper considers Pseudo-Relevance Feedback (PRF) methods for dense retrievers in a resource constrained environment such as that of cheap cloud instances or embedded systems (e.g., smartphones and smartwatches), where memory and CPU are limited and GPUs are not present. For this, we propose a transformer-based PRF method (TPRF), which has a much smaller memory footprint and faster inference time compared to other deep language models that employ PRF mechanisms, with a marginal effectiveness loss. TPRF learns how to effectively combine the relevance feedback signals from dense passage representations. Specifically, TPRF provides a mechanism for modelling relationships and weights between the query and the relevance feedback signals. The method is agnostic to the specific dense representation used and thus can be generally applied to any dense retriever.
Register allocation is one of the most important problems for modern compilers. With a practically unlimited number of user variables and a small number of CPU registers, assigning variables to registers without conflicts is a complex task. This work demonstrates the use of casting the register allocation problem as a graph coloring problem. Using technologies such as PyTorch and OpenAI Gymnasium Environments we will show that a Proximal Policy Optimization model can learn to solve the graph coloring problem. We will also show that the labeling of a graph is critical to the performance of the model by taking the matrix representation of a graph and permuting it. We then test the model's effectiveness on each of these permutations and show that it is not effective when given a relabeling of the same graph. Our main contribution lies in showing the need for label reordering invariant representations of graphs for machine learning models to achieve consistent performance.
Cloud deep learning platforms provide cost-effective deep neural network (DNN) training for customers who lack computation resources. However, cloud systems are often untrustworthy and vulnerable to attackers, leading to growing concerns about model privacy. Recently, researchers have sought to protect data privacy in deep learning by leveraging CPU trusted execution environments (TEEs), which minimize the use of cryptography, but existing works failed to simultaneously utilize the computational resources of GPUs to assist in training and prevent model leakage. This paper presents Tempo, the first cloud-based deep learning system that cooperates with TEE and distributed GPUs for efficient DNN training with model confidentiality preserved. To tackle the challenge of preserving privacy while offloading linear algebraic operations from TEE to GPUs for efficient batch computation, we introduce a customized permutation-based obfuscation algorithm to blind both inputs and model parameters. An optimization mechanism that reduces encryption operations is proposed for faster weight updates during backpropagation to speed up training. We implement Tempo and evaluate it with both training and inference for two prevalent DNNs. Empirical results indicate that Tempo outperforms baselines and offers sufficient privacy protection.
In this paper, we present dSASA (differentiable SASA), an exact geometric method to calculate solvent accessible surface area (SASA) analytically along with atomic derivatives on GPUs. The atoms in a molecule are first assigned to tetrahedra in groups of four atoms by Delaunay tetrahedrization adapted for efficient GPU implementation and the SASA values for atoms and molecules are calculated based on the tetrahedrization information and inclusion-exclusion method. The SASA values from the numerical icosahedral-based method can be reproduced with more than 98% accuracy for both proteins and RNAs. Having been implemented on GPUs and incorporated into the software Amber, we can apply dSASA to implicit solvent molecular dynamics simulations with inclusion of this nonpolar term. The current GPU version of GB/SA simulations has been accelerated up to nearly 20-fold compared to the CPU version, outperforming LCPO, a commonly used, fast algorithm for calculating SASA, as the system size increases. While we focus on the accuracy of the SASA calculations for proteins and nucleic acids, we also demonstrate stable GB/SA MD mini-protein simulations.
Neural vocoders model the raw audio waveform and synthesize high-quality audio, but even the highly efficient ones, like MB-MelGAN and LPCNet, fail to run real-time on a low-end device like a smartglass. A pure digital signal processing (DSP) based vocoder can be implemented via lightweight fast Fourier transforms (FFT), and therefore, is a magnitude faster than any neural vocoder. A DSP vocoder often gets a lower audio quality due to consuming over-smoothed acoustic model predictions of approximate representations for the vocal tract. In this paper, we propose an ultra-lightweight differential DSP (DDSP) vocoder that uses a jointly optimized acoustic model with a DSP vocoder, and learns without an extracted spectral feature for the vocal tract. The model achieves audio quality comparable to neural vocoders with a high average MOS of 4.36 while being efficient as a DSP vocoder. Our C++ implementation, without any hardware-specific optimization, is at 15 MFLOPS, surpasses MB-MelGAN by 340 times in terms of FLOPS, and achieves a vocoder-only RTF of 0.003 and overall RTF of 0.044 while running single-threaded on a 2GHz Intel Xeon CPU.
Speech bandwidth extension (BWE) refers to widening the frequency bandwidth range of speech signals, enhancing the speech quality towards brighter and fuller. This paper proposes a generative adversarial network (GAN) based BWE model with parallel prediction of Amplitude and Phase spectra, named AP-BWE, which achieves both high-quality and efficient wideband speech waveform generation. The proposed AP-BWE generator is entirely based on convolutional neural networks (CNNs). It features a dual-stream architecture with mutual interaction, where the amplitude stream and the phase stream communicate with each other and respectively extend the high-frequency components from the input narrowband amplitude and phase spectra. To improve the naturalness of the extended speech signals, we employ a multi-period discriminator at the waveform level and design a pair of multi-resolution amplitude and phase discriminators at the spectral level, respectively. Experimental results demonstrate that our proposed AP-BWE achieves state-of-the-art performance in terms of speech quality for BWE tasks targeting sampling rates of both 16 kHz and 48 kHz. In terms of generation efficiency, due to the all-convolutional architecture and all-frame-level operations, the proposed AP-BWE can generate 48 kHz waveform samples 292.3 times faster than real-time on a single RTX 4090 GPU and 18.1 times faster than real-time on a single CPU. Notably, to our knowledge, AP-BWE is the first to achieve the direct extension of the high-frequency phase spectrum, which is beneficial for improving the effectiveness of existing BWE methods.
Contactless palmprints are comprised of both global and local discriminative features. Most prior work focuses on extracting global features or local features alone for palmprint matching, whereas this research introduces a novel framework that combines global and local features for enhanced palmprint matching accuracy. Leveraging recent advancements in deep learning, this study integrates a vision transformer (ViT) and a convolutional neural network (CNN) to extract complementary local and global features. Next, a mobile-based, end-to-end palmprint recognition system is developed, referred to as Palm-ID. On top of the ViT and CNN features, Palm-ID incorporates a palmprint enhancement module and efficient dimensionality reduction (for faster matching). Palm-ID balances the trade-off between accuracy and latency, requiring just 18ms to extract a template of size 516 bytes, which can be efficiently searched against a 10,000 palmprint gallery in 0.33ms on an AMD EPYC 7543 32-Core CPU utilizing 128-threads. Cross-database matching protocols and evaluations on large-scale operational datasets demonstrate the robustness of the proposed method, achieving a TAR of 98.06% at FAR=0.01% on a newly collected, time-separated dataset. To show a practical deployment of the end-to-end system, the entire recognition pipeline is embedded within a mobile device for enhanced user privacy and security.
In contemporary times, the increasing complexity of the system poses significant challenges to the reliability, trustworthiness, and security of the SACRES. Key issues include the susceptibility to phenomena such as instantaneous voltage spikes, electromagnetic interference, neutron strikes, and out-of-range temperatures. These factors can induce switch state changes in transistors, resulting in bit-flipping, soft errors, and transient corruption of stored data in memory. The occurrence of soft errors, in turn, may lead to system faults that can propel the system into a hazardous state. Particularly in critical sectors like automotive, avionics, or aerospace, such malfunctions can have real-world implications, potentially causing harm to individuals. This paper introduces a novel fault injector designed to facilitate the monitoring, aggregation, and examination of micro-architectural events. This is achieved by harnessing the microprocessor's PMU and the debugging interface, specifically focusing on ensuring the repeatability of fault injections. The fault injection methodology targets bit-flipping within the memory system, affecting CPU registers and RAM. The outcomes of these fault injections enable a thorough analysis of the impact of soft errors and establish a robust correlation between the identified faults and the essential timing predictability demanded by SACRES.
Beam management (BM) protocols are critical for establishing and maintaining connectivity between network radio nodes and User Equipments (UEs). In Distributed Multiple Input Multiple Output systems (D-MIMO), a number of access points (APs), coordinated by a central processing unit (CPU), serves a number of UEs. At mmWave frequencies, the problem of finding the best AP and beam to serve the UEs is challenging due to a large number of beams that need to be sounded with Downlink (DL) reference signals. The objective of this paper is to investigate whether the best AP/beam can be reliably inferred from sounding only a small subset of beams and leveraging AI/ML for inference of best beam/AP. We use Random Forest (RF), MissForest (MF) and conditional Generative Adversarial Networks (c-GAN) for demonstrating the performance benefits of inference.
The emergence of accurate open large language models (LLMs) has led to a race towards performant quantization techniques which can enable their execution on end-user devices. In this paper, we revisit the problem of "extreme" LLM compression-defined as targeting extremely low bit counts, such as 2 to 3 bits per parameter-from the point of view of classic methods in Multi-Codebook Quantization (MCQ). Our algorithm, called AQLM, generalizes the classic Additive Quantization (AQ) approach for information retrieval to advance the state-of-the-art in LLM compression, via two innovations: 1) learned additive quantization of weight matrices in input-adaptive fashion, and 2) joint optimization of codebook parameters across each transformer blocks. Broadly, AQLM is the first scheme that is Pareto optimal in terms of accuracy-vs-model-size when compressing to less than 3 bits per parameter, and significantly improves upon all known schemes in the extreme compression (2bit) regime. In addition, AQLM is practical: we provide fast GPU and CPU implementations of AQLM for token generation, which enable us to match or outperform optimized FP16 implementations for speed, while executing in a much smaller memory footprint.
A demonstration of a real-time and continuous turn-taking prediction system is presented. The system is based on a voice activity projection (VAP) model, which directly maps dialogue stereo audio to future voice activities. The VAP model includes contrastive predictive coding (CPC) and self-attention transformers, followed by a cross-attention transformer. We examine the effect of the input context audio length and demonstrate that the proposed system can operate in real-time with CPU settings, with minimal performance degradation.
In this paper, we present IntervalMDP.jl, a Julia package for probabilistic analysis of interval Markov Decision Processes (IMDPs). IntervalMDP.jl facilitates the synthesis of optimal strategies and verification of IMDPs against reachability specifications and discounted reward properties. The library supports sparse matrices and is compatible with data formats from common tools for the analysis of probabilistic models, such as PRISM. A key feature of IntervalMDP.jl is that it presents both a multi-threaded CPU and a GPU-accelerated implementation of value iteration algorithms for IMDPs. In particular, IntervalMDP.jl takes advantage of the Julia type system and the inherently parallelizable nature of value iteration to improve the efficiency of performing analysis of IMDPs. On a set of examples, we show that IntervalMDP.jl substantially outperforms existing tools for verification and strategy synthesis for IMDPs in both computation time and memory consumption.
The rapid advancement of energy-efficient parallel ultra-low-power (ULP) ucontrollers units (MCUs) is enabling the development of autonomous nano-sized unmanned aerial vehicles (nano-UAVs). These sub-10cm drones represent the next generation of unobtrusive robotic helpers and ubiquitous smart sensors. However, nano-UAVs face significant power and payload constraints while requiring advanced computing capabilities akin to standard drones, including real-time Machine Learning (ML) performance and the safe co-existence of general-purpose and real-time OSs. Although some advanced parallel ULP MCUs offer the necessary ML computing capabilities within the prescribed power limits, they rely on small main memories (<1MB) and ucontroller-class CPUs with no virtualization or security features, and hence only support simple bare-metal runtimes. In this work, we present Shaheen, a 9mm2 200mW SoC implemented in 22nm FDX technology. Differently from state-of-the-art MCUs, Shaheen integrates a Linux-capable RV64 core, compliant with the v1.0 ratified Hypervisor extension and equipped with timing channel protection, along with a low-cost and low-power memory controller exposing up to 512MB of off-chip low-cost low-power HyperRAM directly to the CPU. At the same time, it integrates a fully programmable energy- and area-efficient multi-core cluster of RV32 cores optimized for general-purpose DSP as well as reduced- and mixed-precision ML. To the best of the authors' knowledge, it is the first silicon prototype of a ULP SoC coupling the RV64 and RV32 cores in a heterogeneous host+accelerator architecture fully based on the RISC-V ISA. We demonstrate the capabilities of the proposed SoC on a wide range of benchmarks relevant to nano-UAV applications. The cluster can deliver up to 90GOp/s and up to 1.8TOp/s/W on 2-bit integer kernels and up to 7.9GFLOp/s and up to 150GFLOp/s/W on 16-bit FP kernels.
Large pre-trained models excel in zero/few-shot learning for language and vision tasks but face challenges in multivariate time series (TS) forecasting due to diverse data characteristics. Consequently, recent research efforts have focused on developing pre-trained TS forecasting models. These models, whether built from scratch or adapted from large language models (LLMs), excel in zero/few-shot forecasting tasks. However, they are limited by slow performance, high computational demands, and neglect of cross-channel and exogenous correlations. To address this, we introduce Tiny Time Mixers (TTM), a compact model (starting from 1M parameters) with effective transfer learning capabilities, trained exclusively on public TS datasets. TTM, based on the light-weight TSMixer architecture, incorporates innovations like adaptive patching, diverse resolution sampling, and resolution prefix tuning to handle pre-training on varied dataset resolutions with minimal model capacity. Additionally, it employs multi-level modeling to capture channel correlations and infuse exogenous signals during fine-tuning. TTM outperforms existing popular benchmarks in zero/few-shot forecasting by (4-40%), while reducing computational requirements significantly. Moreover, TTMs are lightweight and can be executed even on CPU-only machines, enhancing usability and fostering wider adoption in resource-constrained environments. The model weights for reproducibility and research use are available at https://huggingface.co/ibm/ttm-research-r2/, while enterprise-use weights under the Apache license can be accessed as follows: the initial TTM-Q variant at https://huggingface.co/ibm-granite/granite-timeseries-ttm-r1, and the latest variants (TTM-B, TTM-E, TTM-A) weights are available at https://huggingface.co/ibm-granite/granite-timeseries-ttm-r2.
Digital twins are an important technology for advancing mobile communications, specially in use cases that require simultaneously simulating the wireless channel, 3D scenes and machine learning. Aiming at providing a solution to this demand, this work describes a modular co-simulation methodology called CAVIAR. Here, CAVIAR is upgraded to support a message passing library and enable the virtual counterpart of a digital twin system using different 6G-related simulators. The main contributions of this work are the detailed description of different CAVIAR architectures, the implementation of this methodology to assess a 6G use case of UAV-based search and rescue mission (SAR), and the generation of benchmarking data about the computational resource usage. For executing the SAR co-simulation we adopt five open-source solutions: the physical and link level network simulator Sionna, the simulator for autonomous vehicles AirSim, scikit-learn for training a decision tree for MIMO beam selection, Yolov8 for the detection of rescue targets and NATS for message passing. Results for the implemented SAR use case suggest that the methodology can run in a single machine, with the main demanded resources being the CPU processing and the GPU memory.
We present the Locally Adaptive Morphable Model (LAMM), a highly flexible Auto-Encoder (AE) framework for learning to generate and manipulate 3D meshes. We train our architecture following a simple self-supervised training scheme in which input displacements over a set of sparse control vertices are used to overwrite the encoded geometry in order to transform one training sample into another. During inference, our model produces a dense output that adheres locally to the specified sparse geometry while maintaining the overall appearance of the encoded object. This approach results in state-of-the-art performance in both disentangling manipulated geometry and 3D mesh reconstruction. To the best of our knowledge LAMM is the first end-to-end framework that enables direct local control of 3D vertex geometry in a single forward pass. A very efficient computational graph allows our network to train with only a fraction of the memory required by previous methods and run faster during inference, generating 12k vertex meshes at $>$60fps on a single CPU thread. We further leverage local geometry control as a primitive for higher level editing operations and present a set of derivative capabilities such as swapping and sampling object parts. Code and pretrained models can be found at https://github.com/michaeltrs/LAMM.
Resource management for cloud-native microservices has attracted a lot of recent attention. Previous work has shown that machine learning (ML)-driven approaches outperform traditional techniques, such as autoscaling, in terms of both SLA maintenance and resource efficiency. However, ML-driven approaches also face challenges including lengthy data collection processes and limited scalability. We present Ursa, a lightweight resource management system for cloud-native microservices that addresses these challenges. Ursa uses an analytical model that decomposes the end-to-end SLA into per-service SLA, and maps per-service SLA to individual resource allocations per microservice tier. To speed up the exploration process and avoid prolonged SLA violations, Ursa explores each microservice individually, and swiftly stops exploration if latency exceeds its SLA. We evaluate Ursa on a set of representative and end-to-end microservice topologies, including a social network, media service and video processing pipeline, each consisting of multiple classes and priorities of requests with different SLAs, and compare it against two representative ML-driven systems, Sinan and Firm. Compared to these ML-driven approaches, Ursa provides significant advantages: It shortens the data collection process by more than 128x, and its control plane is 43x faster than ML-driven approaches. At the same time, Ursa does not sacrifice resource efficiency or SLAs. During online deployment, Ursa reduces the SLA violation rate by 9.0% up to 49.9%, and reduces CPU allocation by up to 86.2% compared to ML-driven approaches.
Transformer has been adopted to image recognition tasks and shown to outperform CNNs and RNNs while it suffers from high training cost and computational complexity. To address these issues, a hybrid approach has become a recent research trend, which replaces a part of ResNet with an MHSA (Multi-Head Self-Attention). In this paper, we propose a lightweight hybrid model which uses Neural ODE (Ordinary Differential Equation) as a backbone instead of ResNet so that we can increase the number of iterations of building blocks while reusing the same parameters, mitigating the increase in parameter size per iteration. The proposed model is deployed on a modest-sized FPGA device for edge computing. The model is further quantized by QAT (Quantization Aware Training) scheme to reduce FPGA resource utilization while suppressing the accuracy loss. The quantized model achieves 79.68% top-1 accuracy for STL10 dataset that contains 96$\times$96 pixel images. The weights of the feature extraction network are stored on-chip to minimize the memory transfer overhead, allowing faster inference. By eliminating the overhead of memory transfers, inference can be executed seamlessly, leading to accelerated inference. The proposed FPGA implementation accelerates the backbone and MHSA parts by 34.01$\times$, and achieves an overall 9.85$\times$ speedup when taking into account the software pre- and post-processing. The FPGA acceleration leads to 7.10$\times$ better energy efficiency compared to the ARM Cortex-A53 CPU. The proposed lightweight Transformer model is demonstrated on Xilinx ZCU104 board for the image recognition of 96$\times$96 pixel images in this paper and can be applied to different image sizes by modifying the pre-processing layer.