Improved machine translation with INT8 hardware using a novel training method.
problem Training accurate machine translation models with limited hardware precision.
method Convert all Transformer matrix multiplications to 8-bit integer (INT8) without sacrificing accuracy.
result INT8 Transformer models achieve BLEU scores 99.3% to 100% relative to FP32 models.
FrostNet improves INT8 quantization efficiency in mobile networks.
problem The importance of network architecture for optimal INT8 quantization.
method Quantization-aware training (QAT) with StatAssist and GradBoost, hardware-aware NAS.
result FrostNets achieve higher recognition accuracy with comparable latency when quantized.
MSD removes dequantization bottleneck in LLM inference by approximating high-precision activations.
problem Dequantization bottleneck in LLM inference on modern AI accelerators.
method MSD decomposes high-precision activations into multiple low-precision components for direct multiplication with quantized weights.
result MSD avoids INT8-to-BF16 weight conversion, reducing dequantization cycles and HBM traffic.
Efficient Winograd convolution for INT8 networks using RNS.
problem Difficulty in applying Winograd algorithm to low-precision quantized networks.
method Extends Winograd algorithm to Residue Number System (RNS) for efficient INT8 convolution.
result Arithmetic complexity reduction up to 7.03x with performance improvement up to 2.30x-4.69x.
Study finds optimal learning rate schedules for sub-100M quantization-aware training across bit-widths.
problem Optimal learning rate schedules for quantization-aware training depend on bit-width.
method Factorial grid testing over bit-width, warmdown fraction, LR magnitude, model size, and seed.
result INT6 QAT requires a different schedule than higher-precision training, falsifying the primary hypothesis.
APQ jointly optimizes neural architecture, pruning, and quantization for efficient inference.
problem Efficient deep learning inference on resource-constrained hardware.
method Joint optimization of neural architecture, pruning, and quantization policy using a quantization-aware accuracy predictor.
result Joint optimization leads to 2.3% higher ImageNet accuracy with reduced latency and energy consumption.
Paper develops a statistical framework for quantized training of deep neural networks.
problem Lack of theoretical understanding of gradient quantization in FQT.
method Presented a statistical framework for analyzing FQT algorithms, viewing quantized gradient as a stochastic estimator of QAT gradient.
result Developed two novel gradient quantizers with smaller variance than existing per-tensor quantizer.
Degree-Quant improves GNN efficiency by quantizing them without losing accuracy.
problem Efficiency of graph neural networks at inference time.
method Architecturally-agnostic method, Degree-Quant, for quantizing GNNs.
result Degree-Quant trained models perform as well as full-precision models and achieve up to 26% gains.
This study improves Ernie's accuracy for INT8 inference by modifying its training process.
problem Improving the accuracy of pre-trained models like Ernie for low precision inference.
method Integrates a regularizer into the training process to make it more robust to quantization.
result Increased INT8 accuracy for Ernie models.
Generative compression technique reduces neural network size and improves performance on microcontrollers.
problem Memory constraints on microcontrollers limit the size of neural networks, especially for 1x1 pointwise (PW) mixers.
method HYPER-TINYPW uses a shared micro-MLP to generate PW kernels from tiny per-layer codes, reducing memory usage.
result HYPER-TINYPW achieves comparable performance to larger models while being significantly smaller (225 kB vs 1.4 MB).
Bio-inspired neuromorphic hardware is a research direction to approach brain's computational power and energy efficiency. Spiking neural networks (SNN) encode information as sparsely distributed spike trains and employ spike-timing-dependent plasticity (STDP) mechanism for learning. Existing hardware implementations of…
Stealthy hardware Trojan exploits DLA architecture vulnerabilities.
problem Security of DLA deployed on hardware accelerators.
method Input Interception Attack (IIA) exploiting statistical properties of DLA outputs.
result Stealthy Trojan can trigger with some definiteness.
New KWS neural networks improve accuracy and power efficiency.
problem Maximizing accuracy and power efficiency in KWS systems.
method Hardware Aware Training (HAT) for LMU-based neural networks.
result Achieved state-of-the-art accuracy and low parameter counts.
Benchmark for DL inference on embedded HWAs, focusing on autonomous driving.
problem Lack of comprehensive benchmarks for DL hardware.
method Developed a benchmark for inference on embedded HWAs, focusing on autonomous driving. Proposed new granularity, benchmark procedures, and performance indicators.
result Identifies mismatches between HWAs and DL models.
New Ising models improve consensus clustering on specialized hardware.
problem Consensus clustering optimization problems.
method Formulated consensus clustering as Ising models and evaluated on specialized hardware.
result Our Ising models outperform existing techniques on consensus clustering.
DANCE optimizes neural network and accelerator design for faster, more efficient DNN execution.
problem Challenges in optimizing neural network and accelerator design for efficient DNN execution.
method Differentiable approach to co-exploration of accelerator and network architecture design.
result Significantly shorter time to achieve superior accuracy and hardware cost metrics.
Recent breakthroughs in Deep Learning (DL) applications have made DL models a key component in almost every modern computing system. The increased popularity of DL applications deployed on a wide-spectrum of platforms have resulted in a plethora of design challenges related to the constraints introduced by the hardware…
Quantum algorithms for CVaR portfolio optimization face trade-offs between hardware coherence and expressibility.
problem Quantum algorithmic resilience for CVaR portfolio optimization
method WS-QAOA vs. HE-VQNN
result WS-QAOA provides exact theoretical mapping but suffers from hardware decoherence, while HE-VQNN preserves hardware coherence but lacks expressibility.
Method predicts hardware resource usage by control software with guaranteed linear convergence.
problem Predicting time-varying hardware resource availability in control software.
method Path structured multimarginal Schrödinger bridge (MSBP) for learning stochastic resource usage.
result Guaranteed linear convergence to accurate prediction of hardware resource utilization.
Paper proposes a faster method for evaluating DNN hardware and software designs.
problem Reducing time for evaluating different DNN hardware and software designs.
method Using virtual hardware models to estimate DNN performance at the concept phase.
result Up to 92% accuracy in predicting DNN inference processing time.
This paper tackles co-design of neural hardware and software to improve efficiency.
problem Designing efficient deep learning systems that consider both hardware and software optimizations together.
method Developed a constrained Bayesian optimization framework to automatically identify profitable design points in the joint hardware/software design space.
result Improved energy-delay product by 18% (ResNet) and 40% (DQN) over hand-tuned systems.
We derive scaling laws for optimizing neural networks in hardware.
problem Optimizing the large parameter space of neural networks in hardware.
method Analytical derivation of scaling laws for Coordinate Descent optimization.
result Convergence is exponential and scales linearly with the number of neurons.
New method bounds hardware noise without assumptions.
problem Estimating hardware noise without assumptions.
method Machine Learning and Conformal Prediction.
result Theoretical upper bounds of fidelity.
This paper highlights new opportunities for designing large-scale machine learning systems as a consequence of blurring traditional boundaries that have allowed algorithm designers and application-level practitioners to stay -- for the most part -- oblivious to the details of the underlying hardware-level implementatio…
Applying deep neural networks (DNNs) in mobile and safety-critical systems, such as autonomous vehicles, demands a reliable and efficient execution on hardware. Optimized dedicated hardware accelerators are being developed to achieve this. However, the design of efficient and reliable hardware has become increasingly d…
Optimized neural networks for Edge TPU achieve high accuracy in real-time image classification.
problem Designing neural networks for hardware accelerators to achieve optimal performance.
method Hardware-aware neural architecture search and model customization for Edge TPU.
result Improved accuracy-latency tradeoff on Pixel 4's Edge TPU compared to existing models.
With the rising popularity of machine learning and the ever increasing demand for computational power, there is a growing need for hardware optimized implementations of neural networks and other machine learning models. As the technology evolves, it is also plausible that machine learning or artificial intelligence wil…
VegasFlow accelerates complex simulations across various hardware platforms.
problem Complex calculations and simulations requiring high-dimensional integrals.
method Monte Carlo integration techniques using Vegas algorithm and TensorFlow.
result Significantly faster performance on various hardware platforms.
The ever increasing computational cost of Deep Neural Networks (DNN) and the demand for energy efficient hardware for DNN acceleration has made accuracy and hardware cost co-optimization for DNNs tremendously important, especially for edge devices. Owing to the large parameter space and cost of evaluating each paramete…
Paper presents efficient algorithms for convolutional neural networks using Winograd minimal filtering.
problem Resource-efficient implementation of convolutional neural networks.
method Winograd minimal filtering trick applied to M-tap filters (M=3,5,7,9,11) for parallel hardware implementation.
result Approximately 30% reduction in multipliers for fully parallel hardware implementation.
Hardware-accelerated RBM solves large combinatorial problems and integer factorization.
problem Solving large combinatorial optimization and integer factorization problems.
method Logically synthesized RBM architecture, hardware acceleration, and efficient training methods.
result Hardware-accelerated RBM factorizes 16-bit numbers with 10000x speed and 32x power improvements.
Quantum algorithm reduces CVA risk-neutral expectation estimation costs.
problem Reducing Monte Carlo sampling cost for CVA on real quantum hardware.
method Noise-aware quantum workflow combining market calibration, discretisation, and oracle construction.
result CABIQAE achieves lower classical post-processing runtime and more effective error exploitation.
Improves VQAs by balancing classical and quantum training resources.
problem Challenges in trainability and resource costs of VQAs on quantum hardware.
method Adopting HELIA Ansatz and combining classical and quantum methods for gradient estimation and training.
result Achieves higher accuracy and success rates in VQE and improved test accuracy in quantum phase classification.
Pipelined Backpropagation trains large models without batches efficiently.
problem Training large models efficiently on hardware with limited batch sizes.
method Fine-grained Pipelined Backpropagation with Spike Compensation and Linear Weight Prediction.
result Fine-grained Pipelined Backpropagation with a batch size of one matches the accuracy of SGD for multiple networks.
HotNAS reduces AI search time from hundreds of GPU hours to less than 3 GPU hours.
problem High time required for AI solution generation from scratch.
method HotNAS starts from a 'hot' state using pre-trained models and integrates compression during co-search.
result Reduces search time from 200 GPU hours to less than 3 GPU hours.
TensorFlow Probability MCMC toolkit improves MCMC efficiency for modern hardware.
problem Inefficient MCMC algorithms on modern hardware.
method Design and implementation of a new MCMC toolkit for TensorFlow.
result Improved MCMC efficiency on modern hardware.
A scalable DL benchmarking platform for evaluating and comparing models, frameworks, and hardware.
problem Lack of a uniform DL benchmarking platform for evaluating and comparing innovations.
method Identified 10 design features for a DL benchmarking platform, proposed MLModelScope, and implemented as an open-source project.
result Demonstrated how model, hardware, and framework selection affect accuracy and performance under different scenarios.
HarDNN detects and protects CNNs from hardware errors.
problem Hardware errors in CNNs can corrupt inference output.
method Statistical error injection and heuristic vulnerability assessment.
result HarDNN improves CNN resilience with minimal additional computation.
On-device CNN inference for real-time computer vision applications can result in computational demands that far exceed the energy budgets of mobile devices. This paper proposes FixyNN, a co-designed hardware accelerator platform which splits a CNN model into two parts: a set of layers that are fixed in the hardware pla…
Novel NAS method balances performance and hardware metrics efficiently.
problem Challenging multi-objective optimization in neural architecture search.
method Parameterizes joint architectural distribution via hypernetwork conditioned on hardware features and preferences.
result Zero-shot transferability to new devices with representative and diverse architectures.
Galen algorithm compresses neural networks for specific hardware with reduced latency.
problem Finding optimal compression policies for neural networks on specific hardware.
method Reinforcement learning using pruning and quantization to optimize inference latency.
result Compressed ResNet18 for ARM processor reduced inference latency by 80%.
Large-scale deep neural networks are both memory intensive and computation-intensive, thereby posing stringent requirements on the computing platforms. Hardware accelerations of deep neural networks have been extensively investigated in both industry and academia. Specific forms of binary neural networks (BNNs) and sto…
ALF reduces network parameters and operations by 70% and 61%, respectively, on embedded hardware.
problem Efficient deployment of deep learning models on resource-constrained hardware.
method Autoencoder-based low-rank filter-sharing technique.
result ALF achieves significant compression with minimal accuracy loss.
Recurrent neural networks (RNNs) have shown excellent performance in processing sequence data. However, they are both complex and memory intensive due to their recursive nature. These limitations make RNNs difficult to embed on mobile devices requiring real-time processes with limited hardware resources. To address the…
New method uses surrogate gradients to train efficient spiking networks on neuromorphic hardware.
problem Training high-performing spiking networks on analog neuromorphic hardware is challenging due to device mismatch and lack of efficient algorithms.
method Introduces a general in-the-loop learning framework based on surrogate gradients.
result Learning self-corrects for device mismatch, resulting in competitive spiking network performance.
Efficient deep learning computing requires algorithm and hardware co-design to enable specialization: we usually need to change the algorithm to reduce memory footprint and improve energy efficiency. However, the extra degree of freedom from the algorithm makes the design space much larger: it's not only about designin…
Major advancements in building general-purpose and customized hardware have been one of the key enablers of versatility and pervasiveness of machine learning models such as deep neural networks. To sustain this ubiquitous deployment of machine learning models and cope with their computational and storage complexity, se…
NASCaps automates CapsNet design for better accuracy and hardware efficiency.
problem Designing Capsule Networks is laborious and inefficient.
method Automated Neural Architecture Search (NAS) with Genetic Algorithm optimization.
result Jointly optimizes network accuracy and hardware efficiency.