Deep Learning Programming
Build actual neural networks from scratch. This isn't about clicking through tutorials or following pre-made templates. You'll write the code that powers modern AI systems, understand why algorithms work the way they do, and develop the skills Taiwan's tech sector actually needs right now.
What You'll Actually Do
Most courses throw frameworks at you and call it teaching. We start with the math and work up. You'll implement backpropagation by hand before touching TensorFlow. Sounds harder? It is. But when you finish, you'll understand what's happening under the hood.
The program runs for nine months starting September 2025. Three nights per week, plus weekend lab sessions where you tackle real problems. No lectures longer than 45 minutes because nobody learns complex material sitting still for hours.
- Build convolutional networks for image recognition without copying code from Stack Overflow
- Train recurrent networks that actually converge instead of exploding gradients everywhere
- Debug models that aren't working and figure out why your loss function is lying to you
- Work with Taiwan-based datasets including Mandarin language processing challenges
The Curriculum Breakdown
Foundations & Mathematics
Linear algebra, calculus, and probability theory. The stuff you need before anything else makes sense. We cover chain rule until you dream about it.
Neural Networks from Zero
Write your first network in NumPy. No frameworks yet. Just you, matrices, and activation functions. This is where most people realize how much they didn't understand.
Convolutional Architectures
Image recognition, object detection, and why pooling layers matter more than you think. Build a classifier that actually works on messy real-world photos.
Sequential Models & NLP
RNNs, LSTMs, and transformers. Process text data, handle Mandarin character embeddings, and understand attention mechanisms beyond the hype.
Optimization & Debugging
Your models will break. Learn to fix them. Vanishing gradients, overfitting, learning rates that make no sense. We cover the problems nobody talks about in blog posts.
Production Systems
Deploy models that don't crash. Handle edge cases, monitor performance, and write code that other developers can actually work with. This is what separates hobbyists from engineers.
Who's Teaching This
Three engineers who've spent years building ML systems in production. They've made every mistake you're about to make, which means they know how to help you avoid them.
Silje Bakken
Neural Networks LeadEight years building computer vision systems for manufacturing. Previously worked at TSMC developing defect detection models. She'll teach you why your convolutions aren't working.
Vesna Tomlinson
NLP SpecialistBuilt language models for Taiwanese e-commerce platforms. Handles both English and Mandarin processing challenges. She's debugged more broken RNNs than anyone should have to.
Aino Strand
Systems EngineerTakes models from Jupyter notebooks to production environments. Previously at MediaTek optimizing inference speeds for mobile chips. She knows why your training loop is too slow.
How to Join
Technical Assessment
Complete a coding challenge that tests Python fundamentals and basic math. Takes about two hours. We need to know you can handle the material.
Technical Interview
30-minute video call to discuss your background and goals. We're checking if the program fits what you're trying to learn and whether you'll keep up with the pace.
Enrollment Confirmation
Get accepted and secure your spot for the September 2025 cohort. Class size is limited to 24 students so everyone gets actual attention during lab sessions.
Applications Open June 2025
Next cohort begins September 15, 2025. Assessment period runs June through August. If you're not sure about your Python skills, we offer a prep course starting July 2025.
Get Program Details