Efficient Brain-Inspired Hyperdimensional Learning with Spatiotemporal Structured Data

Kim, Jiseung; Lee, Hyunsei; Imani, Mohsen; Kim, Yeseong

doi:10.1109/mascots53633.2021.9614302

Cited by 4 publications

(5 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several works increase the complexity of HDC by using multi-bit representations (i.e., Multi-bit HDC ) instead of single-bit (i.e., binary). Imani et al ( 2019 ), Chuang et al ( 2020 ), Kazemi et al ( 2021 ), Kim et al ( 2021 ), and Yu et al ( 2022 ) use vectors with more complex elements achieving 95.5, 96.6, 98, 98.09, and 98.2%, respectively. With only the latter three achieving slightly higher accuracy than ours, we can conclude that our proposed binary, native HDC method using local linear mapping and POI encoding achieves comparable results with these more complex multi-bit HDC methods.…”

Section: Discussionmentioning

confidence: 99%

An encoding framework for binarized images using hyperdimensional computing

Smets,

Van Leekwijck,

Tsang

et al. 2024

Front. Big Data

View full text Add to dashboard Cite

IntroductionHyperdimensional Computing (HDC) is a brain-inspired and lightweight machine learning method. It has received significant attention in the literature as a candidate to be applied in the wearable Internet of Things, near-sensor artificial intelligence applications, and on-device processing. HDC is computationally less complex than traditional deep learning algorithms and typically achieves moderate to good classification performance. A key aspect that determines the performance of HDC is encoding the input data to the hyperdimensional (HD) space.MethodsThis article proposes a novel lightweight approach relying only on native HD arithmetic vector operations to encode binarized images that preserves the similarity of patterns at nearby locations by using point of interest selection and local linear mapping.ResultsThe method reaches an accuracy of 97.92% on the test set for the MNIST data set and 84.62% for the Fashion-MNIST data set.DiscussionThese results outperform other studies using native HDC with different encoding approaches and are on par with more complex hybrid HDC models and lightweight binarized neural networks. The proposed encoding approach also demonstrates higher robustness to noise and blur compared to the baseline encoding.

show abstract

Section: Discussionmentioning

confidence: 99%

An encoding framework for binarized images using hyperdimensional computing

Smets,

Van Leekwijck,

Tsang

et al. 2024

Front. Big Data

View full text Add to dashboard Cite

show abstract

“…• 2D Position Hypervectors: Drawing inspiration from the work in 60 , we use a set of position hypervectors P 0,0 , P 0,1 , ..., P r−1,c−1 to represent the locations of active events, where r, c indices correspond to the row and column positions in the active pixel sensor frame. These 2D position HD are designed to preserve the spatial correlation between neighbouring events, meaning that if two pixels are physically close in location, their corresponding HD should be highly correlated.…”

Section: Seed Generationmentioning

confidence: 99%

Efficient Event-Based Robotic Grasping Perception using Hyperdimensional Computing

Hassan,

Zou,

Chen

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Grasping is Fundamental in various robotic applications, particularly within industrial contexts. Accurate inference of object properties is a crucial step toward enhancing grasping quality. Dynamic and Active Vision Sensors (DAVIS), increasingly utilized for robotic grasping, offer superior energy efficiency, lower latency, and higher temporal resolution than traditional cameras. However, the data they generate can be complex and noisy, necessitating substantial pre-processing. In response to these challenges, we introduce GraspHD, an innovative end-to-end algorithm that leverages brain-inspired hyperdimensional computing (HDC) to learn about the size and hardness of objects and estimate the grasping force. This novel approach circumvents the need for resource-intensive pre-processing steps, capitalizing on the simplicity and inherent parallelism of HDC operations. Our comprehensive analysis reveals that GraspHD surpasses state-of-the-art approaches in terms of overall classification accuracy. We have also implemented GraspHD on an FPGA to evaluate system efficiency. The results demonstrate that GraspHD operates at a speed 10x faster and offers an energy efficiency 26 times higher than existing learning algorithms, all while maintaining robust performance in noisy environments. These findings underscore the significant potential of GraspHD as a more efficient and effective solution for real-time robotic grasping applications.

show abstract

Section: Base Generationmentioning

confidence: 99%

“…In other words, the events with closer physical distance have a higher correlation. Using techniques introduced in Gallant and Culliton ( 2016 ) and Kim et al ( 2021 ), we generate position hypervectors in three steps: (1) partition events into smaller non-overlapping k × k windows, and (2) generate randomly generated hypervectors for pixels located on the corner of windows. For example, we generate random hypervector for

.…”

Section: Spatial Encodingmentioning

confidence: 99%

“…The number of dimensions to take from each corner hypervectors depends on the relative position of a pixel within the window such that the generated position hypervectors preserve the 2D spatial correlation between events' positions. For an in-depth description of the spatial interpolation, readers are referred to Gallant and Culliton ( 2016 ) and Kim et al ( 2021 ).…”

Section: Spatial Encodingmentioning

confidence: 99%

See 1 more Smart Citation

EventHD: Robust and efficient hyperdimensional learning with neuromorphic sensor

et al. 2022

Self Cite

View full text Add to dashboard Cite

Brain-inspired computing models have shown great potential to outperform today's deep learning solutions in terms of robustness and energy efficiency. Particularly, Hyper-Dimensional Computing (HDC) has shown promising results in enabling efficient and robust cognitive learning. In this study, we exploit HDC as an alternative computational model that mimics important brain functionalities toward high-efficiency and noise-tolerant neuromorphic computing. We present EventHD, an end-to-end learning framework based on HDC for robust, efficient learning from neuromorphic sensors. We first introduce a spatial and temporal encoding scheme to map event-based neuromorphic data into high-dimensional space. Then, we leverage HDC mathematics to support learning and cognitive tasks over encoded data, such as information association and memorization. EventHD also provides a notion of confidence for each prediction, thus enabling self-learning from unlabeled data. We evaluate EventHD efficiency over data collected from Dynamic Vision Sensor (DVS) sensors. Our results indicate that EventHD can provide online learning and cognitive support while operating over raw DVS data without using the costly preprocessing step. In terms of efficiency, EventHD provides 14.2× faster and 19.8× higher energy efficiency than state-of-the-art learning algorithms while improving the computational robustness by 5.9×.

show abstract

Efficient Brain-Inspired Hyperdimensional Learning with Spatiotemporal Structured Data

Cited by 4 publications

References 25 publications

An encoding framework for binarized images using hyperdimensional computing

An encoding framework for binarized images using hyperdimensional computing

Efficient Event-Based Robotic Grasping Perception using Hyperdimensional Computing

EventHD: Robust and efficient hyperdimensional learning with neuromorphic sensor

Contact Info

Product

Resources

About