Task-Projected Hyperdimensional Computing for Multi-task Learning

Chang, Cheng-Yang; Chuang, Yu-Chuan; Wu, An-Yeu

doi:10.1007/978-3-030-49161-1_21

Cited by 13 publications

(9 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It has been shown that orthogonal, high-dimensional context vectors can be bound to neural network weights to access task-specific models stored in superposition with other models [6]. A similar result has also been demonstrated for HD classification, where orthogonal HVs representing different tasks were used as keys to access task-specific prototypes [4]. These strategies could be applicable to EMG-based gesture recognition, with gesture classification in each limb position treated as a separate task.…”

mentioning

confidence: 86%

“…This is in contrast to the strategy proposed in this work, where we approximate the use of different sensor modalities for each stage of dual-stage classification for multiple contexts. Our context-based orthogonalization method is based on work done using orthogonal context vectors to minimize interference between different model parameters stored in superposition [4,6]. Here, we further analyze the benefits of this method when applied to HD computing, and we demonstrate ways to incorporate sensor fusion for creating context-specific HVs using both context classification and direct encoding of accelerometer signals.…”

Section: Related Workmentioning

confidence: 99%

“…However, this method would require separate storage of context-specific parameters and scale memory footprint by the number of desired contexts. We can instead orthogonalize gesture prototypes from different limb position contexts by binding them with orthogonal context vectors prior to superposition [4,6]. During inference, query hypervectors are also bound to their corresponding context vectors before nearest neighbor search.…”

Section: Context-based Orthogonalizationmentioning

confidence: 99%

See 2 more Smart Citations

Memory-Efficient, Limb Position-Aware Hand Gesture Recognition using Hyperdimensional Computing

Zhou,

Muller,

Rabaey

2021

Preprint

View full text Add to dashboard Cite

Electromyogram (EMG) pattern recognition can be used to classify hand gestures and movements for human-machine interface and prosthetics applications, but it often faces reliability issues resulting from limb position change. One method to address this is dual-stage classification, in which the limb position is first determined using additional sensors to select between multiple position-specific gesture classifiers. While improving performance, this also increases model complexity and memory footprint, making a dual-stage classifier difficult to implement in a wearable device with limited resources. In this paper, we present sensor fusion of accelerometer and EMG signals using a hyperdimensional computing model to emulate dualstage classification in a memory-efficient way. We demonstrate two methods of encoding accelerometer features to act as keys for retrieval of position-specific parameters from multiple models stored in superposition. Through validation on a dataset of 13 gestures in 8 limb positions, we obtain a classification accuracy of up to 93.34%, an improvement of 17.79% over using a model trained solely on EMG. We achieve this while only marginally increasing memory footprint over a single limb position model, requiring 8× less memory than a traditional dual-stage classification architecture.

show abstract

mentioning

confidence: 86%

Section: Related Workmentioning

confidence: 99%

Section: Context-based Orthogonalizationmentioning

confidence: 99%

See 1 more Smart Citation

Memory-Efficient, Limb Position-Aware Hand Gesture Recognition using Hyperdimensional Computing

Zhou,

Muller,

Rabaey

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…VSA finds its applications in, for example, cognitive architectures [21], natural language processing [22]- [24], biomedical signal processing [1], [25], approximation of conventional data structures [26], [27], and for classification tasks such as gesture recognition [1], [28], cyber threat detection [29], physical activity recognition [30], fault isolation [31], [32]. Examples of efforts on using VSA for other than classification learning tasks are using data HVs for clustering [33]- [35], semi-supervised learning [36], collaborative privacy-preserving learning [37], [38], multi-task learning [39], [40], distributed learning [41], [42].…”

Section: Related Workmentioning

confidence: 99%

“…The usage of VSA in the multi-task learning context has recently gained an increased attention. The recent works [2], [4], [39], [40], [64] proposed using context hypervectors for multi-task hypervectors in the supervised learning concept. The usage of context hypervectors with Hyperseed is natural.…”

Section: Hyperseed In the Multitask Learning Contextmentioning

confidence: 99%

HyperSeed: Unsupervised Learning with Vector Symbolic Architectures

Osipov¹,

Kahawala²,

Haputhanthri³

et al. 2021

Preprint

View full text Add to dashboard Cite

Motivated by recent innovations in biologicallyinspired neuromorphic hardware, this paper presents a novel unsupervised machine learning approach named Hyperseed that leverages Vector Symbolic Architectures (VSA) for fast learning a topology preserving feature map of unlabelled data. It relies on two major capabilities of VSAs: the binding operation and computing in superposition. In this paper, we introduce the algorithmic part of Hyperseed expressed within Fourier Holographic Reduced Representations VSA model, which is specifically suited for implementation on spiking neuromorphic hardware. The two distinctive novelties of the Hyperseed algorithm are: 1) Learning from only few input data samples and 2) A learning rule based on a single vector operation. These properties are demonstrated on synthetic datasets as well as on illustrative benchmark usecases, IRIS classification and a language identification task using n-gram statistics.

show abstract

On the Vulnerability of Hyperdimensional Computing-Based Classifiers to Adversarial Attacks

Yang

Ren

2020

Network and System Security

View full text Add to dashboard Cite

Task-Projected Hyperdimensional Computing for Multi-task Learning

Cited by 13 publications

References 11 publications

Memory-Efficient, Limb Position-Aware Hand Gesture Recognition using Hyperdimensional Computing

Memory-Efficient, Limb Position-Aware Hand Gesture Recognition using Hyperdimensional Computing

HyperSeed: Unsupervised Learning with Vector Symbolic Architectures

On the Vulnerability of Hyperdimensional Computing-Based Classifiers to Adversarial Attacks

Contact Info

Product

Resources

About