DNNs are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy and analysis quality. Traditional memory saving techniques such as data recomputation and migration either suffers from a high performance overhead or is constrained by specific interconnect technology and limited bandwidth. In this paper, we propose a novel memory-driven high performance CNN training framework that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger neural networks. We evaluate our design against state-of-the-art solutions with four widely-adopted CNNs and the ImangeNet dataset. Results demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5x and 1.8x over the baseline training and state-of-the-art framework with compression, respectively, with little or no accuracy loss. The full paper can be referred to at https://arxiv.org/abs/2011.09017.
more »
« less
COMET: a novel memory-efficient deep learning training framework by using error-bounded lossy compression
Deep neural networks (DNNs) are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy and analysis quality. Training wide and deep neural networks require large amounts of storage resources such as memory because the intermediate activation data must be saved in the memory during forward propagation and then restored for backward propagation. However, state-of-the-art accelerators such as GPUs are only equipped with very limited memory capacities due to hardware design constraints, which significantly limits the maximum batch size and hence performance speedup when training large-scale DNNs. Traditional memory saving techniques either suffer from performance overhead or are constrained by limited interconnect bandwidth or specific interconnect technology. In this paper, we propose a novel memory-efficient CNN training framework (called COMET) that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger models or to accelerate training. Our framework purposely adopts error-bounded lossy compression with a strict error-controlling mechanism. Specifically, we perform a theoretical analysis on the compression error propagation from the altered activation data to the gradients, and empirically investigate the impact of altered gradients over the training process. Based on these analyses, we optimize the error-bounded lossy compression and propose an adaptive error-bound control scheme for activation data compression. Experiments demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5X over the baseline training and 1.8X over another state-of-the-art compression-based framework, respectively, with little or no accuracy loss.
more »
« less
- NSF-PAR ID:
- 10309545
- Date Published:
- Journal Name:
- Proceedings of the VLDB Endowment
- Volume:
- 15
- Issue:
- 4
- ISSN:
- 2150-8097
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Deep neural networks (DNNs) have gained considerable attention in various real-world applications due to the strong performance on representation learning. However, a DNN needs to be trained many epochs for pursuing a higher inference accuracy, which requires storing sequential versions of DNNs and releasing the updated versions to users. As a result, large amounts of storage and network resources are required, which significantly hamper DNN utilization on resource-constrained platforms (e.g., IoT, mobile phone). In this paper, we present a novel delta compression framework called Delta-DNN, which can efficiently compress the float-point numbers in DNNs by exploiting the floats similarity existing in DNNs during training. Specifically, (1) we observe the high similarity of float-point numbers between the neighboring versions of a neural network in training; (2) inspired by delta compression technique, we only record the delta (i.e., the differences) between two neighboring versions, instead of storing the full new version for DNNs; (3) we use the error-bounded lossy compression to compress the delta data for a high compression ratio, where the error bound is strictly assessed by an acceptable loss of DNNs’ inference accuracy; (4) we evaluate Delta-DNN’s performance on two scenarios, including reducing the transmission of releasing DNNs over network and saving the storage space occupied by multiple versions of DNNs. According to experimental results on six popular DNNs, DeltaDNN achieves the compression ratio 2x~10x higher than state-ofthe-art methods, while without sacrificing inference accuracy and changing the neural network structure.more » « less
-
null (Ed.)Nanopore genome sequencing is the key to enabling personalized medicine, global food security, and virus surveillance. The state-of-the-art base-callers adopt deep neural networks (DNNs) to translate electrical signals generated by nanopore sequencers to digital DNA symbols. A DNN-based base-caller consumes 44.5% of total execution time of a nanopore sequencing pipeline. However, it is difficult to quantize a base-caller and build a power-efficient processing-in-memory (PIM) to run the quantized base-caller. Although conventional network quantization techniques reduce the computing overhead of a base-caller by replacing floating-point multiply-accumulations by cheaper fixed-point operations, it significantly increases the number of systematic errors that cannot be corrected by read votes. The power density of prior nonvolatile memory (NVM)-based PIMs has already exceeded memory thermal tolerance even with active heat sinks, because their power efficiency is severely limited by analog-to-digital converters (ADC). Finally, Connectionist Temporal Classification (CTC) decoding and read voting cost 53.7% of total execution time in a quantized base-caller, and thus became its new bottleneck. In this paper, we propose a novel algorithm/architecture co-designed PIM, Helix, to power-efficiently and accurately accelerate nanopore base-calling. From algorithm perspective, we present systematic error aware training to minimize the number of systematic errors in a quantized base-caller. From architecture perspective, we propose a low-power SOT-MRAM-based ADC array to process analog-to-digital conversion operations and improve power efficiency of prior DNN PIMs. Moreover, we revised a traditional NVM-based dot-product engine to accelerate CTC decoding operations, and create a SOT-MRAM binary comparator array to process read voting. Compared to state-of-the-art PIMs, Helix improves base-calling throughput by 6x, throughput per Watt by 11.9x and per mm2 by 7.5x without degrading base-calling accuracy.more » « less
-
Modern deep neural networks (DNNs) often require high memory consumption and large computational loads. In order to deploy DNN algorithms efficiently on edge or mobile devices, a series of DNN compression algorithms have been explored, including factorization methods. Factorization methods approximate the weight matrix of a DNN layer with the multiplication of two or multiple low-rank matrices. However, it is hard to measure the ranks of DNN layers during the training process. Previous works mainly induce low-rank through implicit approximations or via costly singular value decomposition (SVD) process on every training step. The former approach usually induces a high accuracy loss while the latter has a low efficiency. In this work, we propose SVD training, the first method to explicitly achieve low-rank DNNs during training without applying SVD on every step. SVD training first decomposes each layer into the form of its full-rank SVD, then performs training directly on the decomposed weights. We add orthogonality regularization to the singular vectors, which ensure the valid form of SVD and avoid gradient vanishing/exploding. Low-rank is encouraged by applying sparsity-inducing regularizers on the singular values of each layer. Singular value pruning is applied at the end to explicitly reach a low-rank model. We empirically show that SVD training can significantly reduce the rank of DNN layers and achieve higher reduction on computation load under the same accuracy, comparing to not only previous factorization methods but also state-of-the-art filter pruning methods.more » « less
-
Vast volumes of data are produced by today’s scientific simulations and advanced instruments. These data cannot be stored and transferred efficiently because of limited I/O bandwidth, network speed, and storage capacity. Error-bounded lossy compression can be an effective method for addressing these issues: not only can it significantly reduce data size, but it can also control the data distortion based on user-defined error bounds. In practice, many scientific applications have specific requirements or constraints for lossy compression, in order to guarantee that the reconstructed data are valid for post hoc analysis. For example, some datasets contain irrelevant data that should be isolated in particular and users often have intuition regarding value ranges, geospatial regions, and other data subsets that are crucial for subsequent analysis. Existing state-of-the-art error-bounded lossy compressors, however, do not consider these constraints during compression, resulting in inferior compression ratios with respect to user’s post hoc analysis, due to the fact that the data itself provides little or no value for post hoc analysis. In this work we address this issue by proposing an optimized framework that can preserve diverse constraints during the error-bounded lossy compression, e.g., cleaning the irrelevant data, efficiently preserving different precision for multiple value intervals, and allowing users to set diverse precision over both regular and irregular regions. We perform our evaluation on a supercomputer with up to 2,100 cores. Experiments with six real-world applications show that our proposed diverse constraints based error-bounded lossy compressor can obtain a higher visual quality or data fidelity on reconstructed data with the same or even higher compression ratios compared with the traditional state-of-the-art compressor SZ. Our experiments also demonstrate very good scalability in compression performance compared with the I/O throughput of the parallel file system.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.14778/3503585.3503597
