Personal website
I am Gan Wang. I have spent several years on physiological signals and the methods around them: deep learning, machine learning, and signal processing. This site keeps the papers, the MATLAB toolboxes, two EEG analysis projects, a dietary-fiber evidence database, and two iOS apps.
01 — Papers
Wang Gan, Wang Hua-li. The paper reports cross-frequency coupling that differs across wrist movements. DOI 10.16798/j.issn.1003-0530.2016.11.008.
Wang G. and Cerf M. A neural-network decoder that uses temporal and spectral features and outperforms a range of benchmark methods on public brain-computer interface datasets. DOI 10.3389/fninf.2022.952474.
Wang G. Improves parameter estimation inside the common spatial pattern pipeline, tested on public motor-imagery brain-computer interface datasets.
Wang G. Uses a temporal filter-bank common spatial pattern (TFBCSP) to classify EEG recorded during reach and grasp, with higher performance than the comparison methods in the paper.
Wang G. Introduces SSTGCN together with an active-learning procedure for sleep staging, and reports state-of-the-art performance on the datasets used in the paper.
Wang G. Combines a transformer with a convolutional network for sleep scoring, and reports state-of-the-art performance on the datasets used in the paper.
02 — Code and toolboxes
About twenty commonly used nonlinear EEG features, including complexity, entropy, and scaling measures.
About twenty existing variants of non-negative matrix factorization, implemented in MATLAB.
About twenty measures of connectivity (correlation, Granger, and nonlinear) and coupling (coherence, phase, and modulation).
03 — Analysis project
EEG and behavior were recorded while people performed a time-shooting task before and after watching an interesting or a boring movie. The question was whether the movie changed how accurately the brain timed an interval. The analysis covered spectra, event-related potentials, and behavior. The main result: after a boring movie, the error between actual time and felt time grew; after an interesting movie, that difference was not significant. Figures and the MATLAB analysis code are below.
EEG and behavior were recorded from pairs of people playing a Nintendo game or watching videos. The question was whether their brains became more or less aligned as they repeated the same behavior. The archive is the MATLAB analysis written for that project, with a short instruction note.
04 — Dataset
This is a local SQLite knowledge base that ties named dietary fibers to composition, processing, physical behavior, clinical evidence, gut microbes, fermentation, and the papers behind those claims. The database file is not hosted here. If you would like it, contact 296309197@qq.com.
Each fiber has Chinese and English names, a chemical class, and a solubility class. The catalog runs from cereal, fruit, and vegetable fibers through gums, pectins, beta-glucans, resistant starch, and oligosaccharides. Linked tables hold food content, claimed functions, graded clinical evidence, gut microbes and fermentation products, processing, physical and sensory properties, glycemic index, dose, and whether the fiber was chemically modified.
05 — App
A sleep companion that does not need an Apple Watch. Wind down with a large sound library and a short breathing guide, leave the phone on the nightstand, and read a morning report with a quality score and estimated stages. Stage estimates are for wellness only. Sleeppia is not a medical device.
An entertainment app. Parent photos become a generated baby face, then a family picture, a playful similarity score, an ultrasound-style preview, and a short dance clip. The results are for fun. They are not a medical opinion, a genetic test, or a reading of a real ultrasound.