We started teaching neural networks when most people thought they were science fiction. Back in 2019, a handful of us were fascinated by what these algorithms could do—and frustrated by how few resources existed to actually learn them properly.
Three researchers working late nights on deep learning models realized something obvious. The gap between academic papers and practical implementation was enormous. You could read a hundred research papers and still have no idea how to build something that worked.
So we started hosting weekend workshops in Kaohsiung. Just small groups, working through actual code together. People showed up because they wanted to build things, not just understand theory. That hands-on approach became our foundation.
By 2022, those weekend sessions had grown into structured programs. We moved into our current space on Jianguo 1st Road and started developing curriculum that bridges theory with implementation. Not the easiest way to teach, but definitely the most effective.
Our instructors spend their days working with neural networks in production environments. Teaching isn't a side gig—it's how we process what we're learning and share approaches that actually work.
Spent six years optimizing CNN architectures for medical imaging before realizing he wanted to teach. Jasper breaks down complex model designs into digestible pieces. He's particularly good at explaining why certain architectures work better for specific problems—something you won't find in most textbooks.
Built production ML systems for fintech companies across Asia. Dmitri focuses on the messy parts—data preprocessing, model deployment, debugging when things break at 3 AM. His sessions emphasize practical skills over theoretical perfection, which students appreciate when they start working on real projects.
We don't follow the traditional lecture format because it doesn't work well for this material. Neural networks make sense when you're debugging them, tweaking architectures, watching training curves. That requires time with actual code, not PowerPoint slides.
Our programs run for several months because building intuition takes time. You'll work on progressively complex projects, starting with basic feedforward networks and moving toward more sophisticated architectures. We provide computing resources—you don't need expensive hardware to participate.
Every concept gets implemented immediately. We explain the math when it helps understanding, but you'll spend most of your time writing and debugging actual neural network code. That's where real learning happens.
Clean datasets are rare in practice. You'll work with messy data, handle class imbalances, deal with missing values. These are the problems you'll face in actual work, so we build those skills from the start.
Understanding why certain architectures work requires experimenting with variations. You'll modify network structures, compare results, develop intuition about what works for different problems. This experimental approach builds deeper understanding.
Models that work in notebooks often fail in production. We cover deployment challenges, inference optimization, monitoring model performance over time. These practical concerns get addressed throughout the program, not added as an afterthought.
The field moves fast. New architectures appear constantly, research papers pile up, frameworks evolve. We can't teach you everything—but we can help you develop the skills to keep learning and adapting as the technology changes.
Working models beat elegant theory. We emphasize getting things running, then improving them. This iterative approach mirrors how neural network development actually happens in professional settings.
Neural networks aren't magic. They have constraints, require significant data, can fail in unexpected ways. We discuss these limitations openly so you develop realistic expectations about what the technology can and can't do.
Small cohorts work better than large classes. You'll collaborate with other students, review each other's code, discuss different approaches. This peer learning component often provides insights that formal instruction misses.
