Implications of Noise in Resistive Memory on Deep Neural Networks for Image Classification

Resistive memory is a promising alternative to SRAM, but is also an inherently unstable device that requires substantial effort to ensure correct read and write operations. To avoid the associated costs in terms of area, time and energy, the present work is concerned with exploring how much noise in memory operations can be tolerated by image classification tasks based on neural networks. We introduce a special noisy operator that mimics the noise in an exemplary resistive memory unit, explore the resilience of convolutional neural networks on the CIFAR-10 classification task, and discuss a couple of countermeasures to improve this resilience.

Compressing the Backward Pass of Large-Scale Neural Architectures by Structured Activation Pruning

The rise of Deep Neural Networks (DNNs) has led to an increase in model size and complexity, straining the memory capacity of GPUs. Sparsity in DNNs, characterized as structural or ephemeral, has gained attention as a solution. This work focuses on ephemeral sparsity, aiming to reduce memory consumption during training. It emphasizes the significance of activations, an often overlooked component, and their role in memory usage. This work employs structured pruning in Block Sparse Compressed Row (BSR) format in combination with a magnitude-based criterion to efficiently prune activations. We furthermore introduce efficient block-sparse operators for GPUs and showcase their effectiveness, as well as the superior compression offered by block sparsity. We report the effectiveness of activation pruning by evaluating training speed, accuracy, and memory usage of large-scale neural architectures on the example of ResMLP on image classification tasks. As a result, we observe a memory reduction of up to 32% while maintaining accuracy. Ultimately, our approach aims to democratize large-scale model training, reduce GPU requirements, and address ecological concerns.

On the Non-Associativity of Analog Computations

The energy efficiency of analog forms of computing makes it one of the most promising candidates to deploy resource-hungry machine learning tasks on resource-constrained system such as mobile or embedded devices. However, it is well known that for analog computations the safety net of discretization is missing, thus all analog computations are exposed to a variety of imperfections of corresponding implementations. Examples include non-linearities, saturation effect and various forms of noise. In this work, we observe that the ordering of input operands of an analog operation also has an impact on the output result, which essentially makes analog computations non-associative, even though the underlying operation might be mathematically associative. We conduct a simple test by creating a model of a real analog processor which captures such ordering effects. With this model we assess the importance of ordering by comparing the test accuracy of a neural network for keyword spotting, which is trained based either on an ordered model, on a non-ordered variant, and on real hardware. The results prove the existence of ordering effects as well as their high impact, as neglecting ordering results in substantial accuracy drops.

Reducing Memory Requirements for the IPU using Butterfly Factorizations

High Performance Computing (HPC) benefits from different improvements during last decades, specially in terms of hardware platforms to provide more processing power while maintaining the power consumption at a reasonable level. The Intelligence Processing Unit (IPU) is a new type of massively parallel processor, designed to speedup parallel computations with huge number of processing cores and on-chip memory components connected with high-speed fabrics. IPUs mainly target machine learning applications, however, due to the architectural differences between GPUs and IPUs, especially significantly less memory capacity on an IPU, methods for reducing model size by sparsification have to be considered. Butterfly factorizations are well-known replacements for fully-connected and convolutional layers. In this paper, we examine how butterfly structures can be implemented on an IPU and study their behavior and performance compared to a GPU. Experimental results indicate that these methods can provide 98.5% compression ratio to decrease the immense need for memory, the IPU implementation can benefit from 1.3x and 1.6x performance improvement for butterfly and pixelated butterfly, respectively. We also reach to 1.62x training time speedup on a real-word dataset such as CIFAR10.

Walking Noise: Understanding Implications of Noisy Computations on Classification Tasks

Machine learning methods like neural networks are extremely successful and popular in a variety of applications, however, they come at substantial computational costs, accompanied by high energy demands. In contrast, hardware capabilities are limited and there is evidence that technology scaling is stuttering, therefore, new approaches to meet the performance demands of increasingly complex model architectures are required. As an unsafe optimization, noisy computations are more energy efficient, and given a fixed power budget also more time efficient. However, any kind of unsafe optimization requires counter measures to ensure functionally correct results. This work considers noisy computations in an abstract form, and gears to understand the implications of such noise on the accuracy of neural-network-based classifiers as an exemplary workload. We propose a methodology called "Walking Noise" that allows to assess the robustness of different layers of deep architectures by means of a so-called "midpoint noise level" metric. We then investigate the implications of additive and multiplicative noise for different classification tasks and model architectures, with and without batch normalization. While noisy training significantly increases robustness for both noise types, we observe a clear trend to increase weights and thus increase the signal-to-noise ratio for additive noise injection. For the multiplicative case, we find that some networks, with suitably simple tasks, automatically learn an internal binary representation, hence becoming extremely robust. Overall this work proposes a method to measure the layer-specific robustness and shares first insights on how networks learn to compensate injected noise, and thus, contributes to understand robustness against noisy computations.

Towards Hardware-Specific Automatic Compression of Neural Networks

Compressing neural network architectures is important to allow the deployment of models to embedded or mobile devices, and pruning and quantization are the major approaches to compress neural networks nowadays. Both methods benefit when compression parameters are selected specifically for each layer. Finding good combinations of compression parameters, so-called compression policies, is hard as the problem spans an exponentially large search space. Effective compression policies consider the influence of the specific hardware architecture on the used compression methods. We propose an algorithmic framework called Galen to search such policies using reinforcement learning utilizing pruning and quantization, thus providing automatic compression for neural networks. Contrary to other approaches we use inference latency measured on the target hardware device as an optimization goal. With that, the framework supports the compression of models specific to a given hardware target. We validate our approach using three different reinforcement learning agents for pruning, quantization and joint pruning and quantization. Besides proving the functionality of our approach we were able to compress a ResNet18 for CIFAR-10, on an embedded ARM processor, to 20% of the original inference latency without significant loss of accuracy. Moreover, we can demonstrate that a joint search and compression using pruning and quantization is superior to an individual search for policies using a single compression method.

HW-Aware Initialization of DNN Auto-Tuning to Improve Exploration Time and Robustness

The process of optimizing the latency of DNN operators with ML models and hardware-in-the-loop, called auto-tuning, has established itself as a pervasive method for the deployment of neural networks. From a search space of loop-optimizations, the candidate providing the best performance has to be selected. Performance of individual configurations is evaluated through hardware measurements. The combinatorial explosion of possible configurations, together with the cost of hardware evaluation makes exhaustive explorations of the search space infeasible in practice. Machine Learning methods, like random forests or reinforcement learning are used to aid in the selection of candidates for hardware evaluation. For general purpose hardware like x86 and GPGPU architectures impressive performance gains can be achieved, compared to hand-optimized libraries like cuDNN. The method is also useful in the space of hardware accelerators with less wide-spread adoption, where a high-performance library is not always available. However, hardware accelerators are often less flexible with respect to their programming which leads to operator configurations not executable on the hardware target. This work evaluates how these invalid configurations affect the auto-tuning process and its underlying performance prediction model for the VTA hardware. From these results, a validity-driven initialization method for AutoTVM is developed, only requiring 41.6% of the necessary hardware measurements to find the best solution, while improving search robustness.

The Programming of Deep Learning Accelerators as a Constraint Satisfaction Problem

The success of Deep Artificial Neural Networks (DNNs) in many domains created a rich body of research concerned with hardware accelerators for compute-intensive DNN operators. However, implementing such operators efficiently with complex instructions such as matrix multiply is a task not yet automated gracefully. Solving this task often requires complex program and memory layout transformations. First solutions to this problem have been proposed, such as TVM or ISAMIR, which work on a loop-level representation of operators and rewrite the program before an instruction embedding into the operator is performed. This top-down approach creates a tension between exploration range and search space complexity. In this work, we propose a new approach to this problem. We have created a bottom-up method that allows the direct generation of implementations based on an accelerator's instruction set. By formulating the embedding as a constraint satisfaction problem over the scalar dataflow, every possible embedding solution is contained in the search space. By adding additional constraints, a solver can produce the subset of preferable solutions. A detailed evaluation using the VTA hardware accelerator with the Baidu DeepBench inference benchmark suite shows that our approach can automatically generate code competitive to reference implementations, and furthermore that memory layout flexibilty can be beneficial for overall performance. While the reference implementation achieves very low hardware utilization due to its fixed embedding strategy, we achieve a geomean speedup of up to x2.49, while individual operators can improve as much as x238.

Understanding Cache Boundness of ML Operators on ARM Processors

Machine Learning compilers like TVM allow a fast and flexible deployment on embedded CPUs. This enables the use of non-standard operators, which are common in ML compression techniques. However, it is necessary to understand the limitations of typical compute-intense operators in ML workloads to design a proper solution. This is the first in-detail analysis of dense and convolution operators, generated with TVM, that compares to the fundamental hardware limits of embedded ARM processors. Thereby it explains the gap between computational peak performance, theoretical and measured, and real-world state-of-the-art results, created with TVM and openBLAS. Instead, one can see that single-precision general matrix multiply (GEMM) and convolutions are bound by L1-cache-read bandwidth. Explorations of 8-bit and bit-serial quantized operators show that quantization can be used to achieve relevant speedups compared to cache-bound floating-point operators. However, the performance of quantized operators highly depends on the interaction between data layout and bit packing.