An official website of the United States government
Processing large amounts of data, especially in learning algorithms, poses a challenge for current embedded computing systems. Hyperdimensional (HD) computing (HDC) is a brain-inspired computing paradigm that works with high-dimensional vectors called hypervectors . HDC replaces several complex learning computations with bitwise and simpler arithmetic operations at the expense of an increased amount of data due to mapping the data into high-dimensional space. These hypervectors, more often than not, cannot be stored in memory, resulting in long data transfers from storage. In this article, we propose Store-n-Learn, an in-storage computing solution that performs HDC classification and clustering by implementing encoding, training, retraining, and inference across the flash hierarchy. To hide the latency of training and enable efficient computation, we introduce the concept of batching in HDC. We also present on-chip acceleration for HDC encoding in flash planes. This enables us to exploit the high parallelism provided by the flash hierarchy and encode multiple data points in parallel in both batched and non-batched fashion. Store-n-Learn also implements a single top-level FPGA accelerator with novel implementations for HDC classification training, retraining, inference, and clustering on the encoded data. Our evaluation over 10 popular datasets shows that Store-n-Learn is on average 222× (543×) faster than CPU and 10.6× (7.3×) faster than the state-of-the-art in-storage computing solution, INSIDER for HDC classification (clustering).
more »
« less
Overparameterized neural networks implement associative memory
Identifying computational mechanisms for memorization and retrieval of data is a long-standing problem at the intersection of machine learning and neuroscience. Our main finding is that standard overparameterized deep neural networks trained using standard optimization methods implement such a mechanism for real-valued data. We provide empirical evidence that 1) overparameterized autoencoders store training samples as attractors and thus iterating the learned map leads to sample recovery, and that 2) the same mechanism allows for encoding sequences of examples and serves as an even more efficient mechanism for memory than autoencoding. Theoretically, we prove that when trained on a single example, autoencoders store the example as an attractor. Lastly, by treating a sequence encoder as a composition of maps, we prove that sequence encoding provides a more efficient mechanism for memory than autoencoding.
more »
« less
- PAR ID:
- 10232184
- Date Published:
- Journal Name:
- Proceedings of the National Academy of Sciences
- Volume:
- 117
- Issue:
- 44
- ISSN:
- 0027-8424
- Page Range / eLocation ID:
- 27162 to 27170
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Autoencoders represent a significant category of deep learning models and are widely utilized for dimensionality reduction. However, standard Autoencoders are complicated architectures that normally have several layers and many hyper-parameters that require tuning. In this paper, we introduce a new type of autoencoder that we call dynamical system autoencoder (DSAE). Similar to classic autoencoders, DSAEs can effectively handle dimensionality reduction and denoising tasks, and they demonstrate strong performance in several benchmark tasks. However, DSAEs, in some sense, have a more flexible architecture than standard AEs. In particular, in this paper we study simple DSAEs that only have a single layer. In addition, DSAEs provide several theoretical and practical advantages arising from their implementation as iterative maps, which have been well studied over several decades. Beyond the inherent simplicity of DSAEs, we also demonstrate how to use sparse matrices to reduce the number of parameters for DSAEs without sacrificing the performance of our methods. Our simulation studies indicate that DSAEs achieved better performance than the classic autoencoders when the encoding dimension or training sample size was small. Additionally, we illustrate how to use DSAEs, and denoising autoencoders in general. to nerform sunervised learning tasks.more » « less
-
The colored de Bruijn graph (cdbg) and its variants have become an important combinatorial structure used in numerous areas in genomics, such as population-level variation detection in metagenomic samples, large scale sequence search, and cdbg-based reference sequence indices. As samples or genomes are added to the cdbg, the color information comes to dominate the space required to represent this data structure. In this paper, we show how to represent the color information efficiently by adopting a hierarchical encoding that exploits correlations among color classes—patterns of color occurrence—present in the de Bruijn graph (dbg). A major challenge in deriving an efficient encoding of the color information that takes advantage of such correlations is determining which color classes are close to each other in the high-dimensional space of possible color patterns. We demonstrate that the dbg itself can be used as an efficient mechanism to search for approximate nearest neighbors in this space. While our approach reduces the encoding size of the color information even for relatively small cdbgs (hundreds of experiments), the gains are particularly consequential as the number of potential colors (i.e. samples or references) grows into the thousands. We apply this encoding in the context of two different applications; the implicit cdbg used for a large-scale sequence search index, Mantis, as well as the encoding of color information used in population-level variation detection by tools such as Vari and Rainbowfish. Our results show significant improvements in the overall size and scalability of representation of the color information. In our experiment on 10,000 samples, we achieved more than 11× better compression compared to RRR.more » « less
-
Today's high-performance computing (HPC) applications are producing vast volumes of data, which are challenging to store and transfer efficiently during the execution, such that data compression is becoming a critical technique to mitigate the storage burden and data movement cost. Huffman coding is arguably the most efficient Entropy coding algorithm in information theory, such that it could be found as a fundamental step in many modern compression algorithms such as DEFLATE. On the other hand, today's HPC applications are more and more relying on the accelerators such as GPU on supercomputers, while Huffman encoding suffers from low throughput on GPUs, resulting in a significant bottleneck in the entire data processing. In this paper, we propose and implement an efficient Huffman encoding approach based on modern GPU architectures, which addresses two key challenges: (1) how to parallelize the entire Huffman encoding algorithm, including codebook construction, and (2) how to fully utilize the high memory-bandwidth feature of modern GPU architectures. The detailed contribution is four-fold. (1) We develop an efficient parallel codebook construction on GPUs that scales effectively with the number of input symbols. (2) We propose a novel reduction based encoding scheme that can efficiently merge the codewords on GPUs. (3) We optimize the overall GPU performance by leveraging the state-of-the-art CUDA APIs such as Cooperative Groups. (4) We evaluate our Huffman encoder thoroughly using six real-world application datasets on two advanced GPUs and compare with our implemented multi-threaded Huffman encoder. Experiments show that our solution can improve the encoding throughput by up to 5.0x and 6.8x on NVIDIA RTX 5000 and V100, respectively, over the state-of-the-art GPU Huffman encoder, and by up to 3.3x over the multi-thread encoder on two 28-core Xeon Platinum 8280 CPUs.more » « less
-
Abstract Sequence memory is an essential attribute of natural and artificial intelligence that enables agents to encode, store, and retrieve complex sequences of stimuli and actions. Computational models of sequence memory have been proposed where recurrent Hopfield-like neural networks are trained with temporally asymmetric Hebbian rules. However, these networks suffer from limited sequence capacity (maximal length of the stored sequence) due to interference between the memories. Inspired by recent work on Dense Associative Memories, we expand the sequence capacity of these models by introducing a nonlinear interaction term, enhancing separation between the patterns. We derive novel scaling laws for sequence capacity with respect to network size, significantly outperforming existing scaling laws for models based on traditional Hopfield networks, and verify these theoretical results with numerical simulation. Moreover, we introduce a generalized pseudoinverse rule to recall sequences of highly correlated patterns. Finally, we extend this model to store sequences with variable timing between states’ transitions and describe a biologically-plausible implementation, with connections to motor neuroscience.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1073/PNAS.2005013117
