Yantong Zhu

• Established standardized benchmarking framework for evaluating calcium imaging analysis pipelines on large-scale zebrafish datasets.
• Developed and refined 'Voluseg,' an automated pipeline optimized for volumetric calcium imaging data handling.
• Compared Voluseg, Suite2p, and CaImAn pipelines on volumetric datasets, fine-tuning parameters for enhanced accuracy.
• Implemented preprocessing steps such as detrending, baseline correction, and Gaussian convolution of swim signals.
• Conducted regression analyses to correlate neuronal activity with electrophysiological signals, quantifying physiological relevance.
• Demonstrated Voluseg's robustness in managing volumetric datasets, outperforming Suite2p and CaImAn across depths.
• Navigated complex computational challenges, effectively managing memory and package dependencies in high-performance environments.

• Designed pulse stimulation patterns for non-human primates to analyze motor circuit responses and movements.
• Summarized findings from related research papers to inform experimental design.
• Developed safe stimulation protocols for human research applications.
• Investigated motor evoked potentials post-stimulation to assess intervention effectiveness.

• Designed and patented an innovative transcranial stimulation device for major depressive disorder niche markets.
• Engineered coil design with out-of-phase currents to maximize therapeutic effect.
• Utilized ferrite core electromagnets powered by 9A circuits to enhance performance.
• Developed circuit design, converting AC to DC with voltage step-down and feedback load.

Developed deep learning pipeline for colorectal tissue classification using CNN and ResNet architectures.

• Trained custom 11-layer CNN and fine-tuned Inception_v3 on dataset of 5,000 images across eight classes.
• Achieved 89.9% test accuracy with Inception_v3, surpassing baseline CNN performance.
• Applied sliding-window technique to segment large histology images, creating visual maps of tissue distributions.
• Utilized Python, TensorFlow, Keras, and scikit-learn throughout development process.

• Established a model for precise classification of psychological stress using galvanic skin response data.
• Utilized driver galvanic skin response database from MIT experiments for pre-processing and feature extraction.
• Developed SVM algorithm in Python for effective psychological stress recognition.
• Refined model by adjusting parameters to enhance recognition rate of psychological stress classifier.
• Achieved ideal recognition rates with SVM algorithm for single-mode skin electrical signals.
• Demonstrated low computational costs and simplicity of the algorithm model, indicating future practical significance.

• Explored attention patterns during science problem-solving using eye-tracking techniques.
• Recruited participants and provided guidance on eye-tracking device usage.
• Collected data focusing on Saccade, Fixation, and Areas of Interest (AOIs) parameters.
• Analyzed results indicating students spent more time on relevant factors compared to irrelevant ones.

• Conducted EEG-based emotion classification through signal preprocessing with band-pass filters and manual feature extraction.
• Implemented machine learning models including KNN, Naïve Bayes, and Decision Trees, achieving optimal performance with KNN.
• Developed a simple CNN and assessed its performance against conventional models, noting underperformance due to data limitations.
• Utilized CNN as a feature extractor to train hybrid models (CNN+SVM, CNN+KNN), enhancing classification accuracy compared to standalone CNN.