This repository is part of my individual project for the L46 module (Principles of Machine Learning Systems) in my MPhil in Advanced Computer Science at the University of Cambridge.
This repository contains Hype, a hyperdimensional computing framework built on the standard library, and HDVR, an application of Hype on the ISOLET dataset for speech recognition.
This implementation is based largely on work by Imani et al. (2017):
Imani, M., Kong, D., Rahimi, A. & Simunic, T. (2017), ‘VoiceHD: Hyperdimensional Computing for Efficient Speech Recognition’, 2017 IEEE International Conference on Rebooting Computing (ICRC) pp. 1–8.
The paper can be found on IEEE Xplore.
The core library, Hype, contains general constructions which are commonly used in hyperdimensional computing. These
include interfaces and implementations for hypervectors (see hype::IVector and hype::Vector) and memory (see
hype::Memory and hype::AssociativeMemory).
To demonstrate the core library, it has been used in HDVR. This is a concrete application on the ISOLET dataset for speech recognition.
More information on the project and its performance can be found in the report in this repository.
To compile the example application, run the following commands in a terminal in the root directory:
mkdir build
cd build
cmake ..
make
You can then run the example application by running:
./HDVR
Loading raw datasets... DONE
Encoding training data... 9% 19% 29% 39% 49% 59% 69% 79% 89% 99% DONE
Encoding testing data... 9% 19% 29% 39% 49% 59% 69% 79% 89% 99% DONE
Loaded 6238 training samples, and 1559 testing samples.
Running test...
Accuracy before training: 88.5183%
[Epoch: 1]: error: 8.75281% – accuracy: 92.0462%
[Epoch: 2]: error: 5.96345% – accuracy: 93.2008%
[Epoch: 3]: error: 4.21609% – accuracy: 93.7139%
[Epoch: 4]: error: 2.99776% – accuracy: 93.7139%
[Epoch: 5]: error: 2.35652% – accuracy: 93.9705%
[Epoch: 6]: error: 1.69926% – accuracy: 94.0988%
[Epoch: 7]: error: 1.47483% – accuracy: 94.2912%
[Epoch: 8]: error: 1.042% – accuracy: 94.0346%
[Epoch: 9]: error: 1.00994% – accuracy: 94.0346%
[Epoch: 10]: error: 0.81757% – accuracy: 94.0988%
=== SUCCESS ===