DPQ-HD: Post-Training Compression for Ultra-Low Power Hyperdimensional Computing
Nilesh Prasad Pandey, Shriniwas Kulkarni, David Wang, Onat Gungor, Flavio Ponzina, Tajana Rosing
TL;DR
DPQ-HD tackles the memory and compute bottlenecks of Hyperdimensional Computing on resource-constrained edges by introducing a post-training, end-to-end compression pipeline that jointly applies low-rank projection decomposition, pruning, and quantization in the order $D \rightarrow P \rightarrow Q$, and augments this with an adaptive online inference strategy based on cosine margins. The framework enables retraining-free optimization while maintaining near-floating-point accuracy, achieving up to $20$-$100$× memory reductions and significant MCU speedups (e.g., up to $56$-fold) with only about $1$-$2$% accuracy loss, and reducing offline optimization time by up to $100$× compared to retraining-based SOTA methods. A theoretical justification shows pruning before quantization does not incur excess error beyond the sum of the individual steps, while calibration-based rank/pruning selection adapts to task difficulty. Empirically, DPQ-HD outperforms existing post-training baselines and approaches the performance of retraining-based SOTA, enabling fast, energy-efficient edge deployment of HDC across image, graph, and speech tasks on microcontrollers.
Abstract
Hyperdimensional Computing (HDC) is emerging as a promising approach for edge AI, offering a balance between accuracy and efficiency. However, current HDC-based applications often rely on high-precision models and/or encoding matrices to achieve competitive performance, which imposes significant computational and memory demands, especially for ultra-low power devices. While recent efforts use techniques like precision reduction and pruning to increase the efficiency, most require retraining to maintain performance, making them expensive and impractical. To address this issue, we propose a novel Post Training Compression algorithm, Decomposition-Pruning-Quantization (DPQ-HD), which aims at compressing the end-to-end HDC system, achieving near floating point performance without the need of retraining. DPQ-HD reduces computational and memory overhead by uniquely combining the above three compression techniques and efficiently adapts to hardware constraints. Additionally, we introduce an energy-efficient inference approach that progressively evaluates similarity scores such as cosine similarity and performs early exit to reduce the computation, accelerating prediction inference while maintaining accuracy. We demonstrate that DPQ-HD achieves up to 20-100x reduction in memory for image and graph classification tasks with only a 1-2% drop in accuracy compared to uncompressed workloads. Lastly, we show that DPQ-HD outperforms the existing post-training compression methods and performs better or at par with retraining-based state-of-the-art techniques, requiring significantly less overall optimization time (up to 100x) and faster inference (up to 56x) on a microcontroller