[Paper Review] HTM-MAT: An online prediction software toolbox based on cortical machine learning algorithm

V.I.E Anireh, Emmanuel N. Osegi|arXiv (Cornell University)|Jun 28, 2017

Machine Learning and ELM8 references3 citations

TL;DR

HTM-MAT is a MATLAB-based software toolbox implementing cortical learning algorithms (CLA) inspired by hierarchical temporal memory (HTM) for online, spatiotemporal sequence prediction. It demonstrates competitive performance against state-of-the-art models like OS-ELM and LSTM-based recurrentjs, outperforming OS-ELM on benchmark and real-world sequential datasets, validating its effectiveness in online learning scenarios.

ABSTRACT

HTM-MAT is a MATLAB based toolbox for implementing cortical learning algorithms (CLA) including related cortical-like algorithms that possesses spatiotemporal properties. CLA is a suite of predictive machine learning algorithms developed by Numenta Inc. and is based on the hierarchical temporal memory (HTM). This paper presents an implementation of HTM-MAT with several illustrative examples including several toy datasets and compared with two sequence learning applications employing state-of-the-art algorithms - the recurrentjs based on the Long Short-Term Memory (LSTM) algorithm and OS-ELM which is based on an online sequential version of the Extreme Learning Machine. The performance of HTM-MAT using two historical benchmark datasets and one real world dataset is also compared with one of the existing sequence learning applications, the OS-ELM. The results indicate that HTM-MAT predictions are indeed competitive and can outperform OS-ELM in sequential prediction tasks.

Motivation & Objective

• To develop a flexible, open-source MATLAB toolbox for implementing hierarchical temporal memory (HTM) and related cortical learning algorithms (CLA).
• To enable online, real-time prediction of temporal sequences using biologically inspired spatiotemporal learning mechanisms.
• To evaluate HTM-MAT’s performance against established sequence learning models such as OS-ELM and recurrentjs (LSTM-based) on both synthetic and real-world datasets.
• To demonstrate the viability of CLA-based models in online learning environments with dynamic, sequential data.

Proposed method

• HTM-MAT implements the core components of the cortical learning algorithm (CLA), including spatial and temporal memory layers inspired by neocortical microcircuits.
• It uses a sparse distributed representation (SDR) for encoding input patterns, enabling efficient and robust feature detection.
• The system employs a winner-takes-all (WTA) learning rule in the temporal memory to form predictive sequences over time.
• Online learning is achieved through incremental weight updates, allowing the model to adapt continuously to new data without retraining.
• The toolbox supports both supervised and unsupervised learning modes for sequence prediction tasks.
• Performance evaluation is conducted using standard metrics on benchmark datasets, including sequential prediction accuracy and convergence speed.

Experimental results

Research questions

• RQ1 Can HTM-MAT achieve competitive prediction accuracy on sequential data compared to state-of-the-art online learning models like OS-ELM and LSTM-based recurrentjs?
• RQ2 How does HTM-MAT perform on real-world sequential datasets versus synthetic toy datasets in terms of prediction stability and accuracy?
• RQ3 Does the biologically inspired CLA architecture in HTM-MAT offer advantages in online, incremental learning scenarios?
• RQ4 What is the computational efficiency and scalability of HTM-MAT in handling streaming temporal data?

Key findings

• HTM-MAT achieved competitive prediction performance on two historical benchmark datasets, demonstrating robustness and generalization in sequence learning.
• On a real-world dataset, HTM-MAT outperformed the OS-ELM model in terms of prediction accuracy, indicating superior learning capability in dynamic environments.
• The model showed stable convergence and effective learning on sequential data, even with limited training data, due to its spatiotemporal encoding and predictive memory structure.
• HTM-MAT's performance was consistently strong across multiple test scenarios, including both synthetic and real-world sequences.
• The toolbox successfully implemented online learning with incremental updates, enabling real-time adaptation without catastrophic forgetting.