School of Engineering Department of Electronic and Computer Engineering 150 Neural Coding for Motor Brain Machine Interfaces Supervisor: WANG Yiwen / ECE Student: TAN Tianshu / CPEG Course: UROP 1100, Summer This report explores low-dimensional neural data alignment methods crucial for comparing complex neural dynamics across various conditions. We review techniques for latent space construction and alignment, including Distribution Alignment Decoding and Hierarchical Wasserstein Alignment. The potential of graphbased methods, which preserve neural network topology and offer noise robustness, is also discussed. Building on this review, we present a novel graph-based alignment methodology applied to neural recordings from subjects performing two related motor tasks. Our approach constructs k-nearest neighbor graphs to represent low-dimensional neural data and implements an improved matching and alignment procedure. Preliminary results on a simplified dataset demonstrate the method's ability to align shared neural patterns while preserving local data relationships. This work aims to enhance our understanding of neural transfer learning mechanisms and improve decoding efficiency for brain-computer interfaces. Future directions include refining graph construction methods, developing sophisticated alignment techniques for high-dimensional neural data, and implementing rigorous validation procedures. Design and Characterization of a Tactile Sensor Array Supervisor: WONG Man / ECE Student: HONG Guangbin / ELEC Course: UROP 1100, Summer This project focuses on how to construct a simple spiking neural network using Python to identify written figures. A 2-layer framework is constructed to achieve the target. The SNN models capture iconic features from biological neurons to facilitate recognition, allowing the layers to perform the designated tasks on track. After using specially designed algorithm based on biological neural learning pattern to train the model, the accuracy could reach a maximum of 60%. There exist much room for improvement in aspects of accuracy and stability for the model. Potentials of increasing these by improving the training efficiency, parameters’ adjusting frequency and model design still worth further exploration.
RkJQdWJsaXNoZXIy NDk5Njg=