Neural networks explained through architecture
A structured lecture series on how modern neural networks are built, why specific designs outperform others, and where the field is heading.
A place people recognise themselves
Most learners arrive here after hitting the same wall: tutorials that show what to type but never explain why one layer connects to the next that way. The gap between running someone else's notebook and understanding the design decisions behind it is exactly what these lectures address.
The material assumes familiarity with Python and basic linear algebra — nothing beyond that. Architectural reasoning is built up incrementally, starting from perceptrons and moving through convolutional, recurrent, and attention-based designs with the same analytical lens applied to each.
"I had been reading papers for months without a clear mental model of how the pieces fit together. After the third module on transformer internals I finally saw the pattern — and it changed how I read everything else."
Lecture delivery — module 7, attention mechanisms
What stays once the lectures are done
Architectural literacy is durable. Specific frameworks change quickly; the reasoning behind design choices does not.
The aim is not to produce someone who can reproduce a transformer from memory. The aim is to produce someone who, faced with a new architecture paper, can read the diagram, identify the design decisions, and form an opinion on the trade-offs involved.
That kind of reading fluency takes deliberate exposure to multiple families of networks — CNNs, RNNs, graph networks, diffusion models — examined under the same analytical framework. Students who go through the full series consistently report that research papers feel less opaque afterward.
Architecture types examined with comparative analysis across modules
Typical completion time at 8–10 hours per week of focused study
Written design exercises — one per module, reviewed with written feedback
Access to all recorded lectures and updated materials after enrolment
Authored modules on graph neural networks and diffusion models. Previously a research engineer at two EU-based AI labs.
What the commitment looks like
The series is offered in two access configurations. Both give full video access and all written materials. The difference is in the feedback loop — how much direct interaction with instructors is included.
There are no upsells after enrolment. The price you see covers everything described. Adex Krost does not charge separately for material updates or Q&A access within the same tier.
Full lecture library, all design briefs, and written answer keys. Suitable for learners who prefer to work independently and do not require instructor review of their work.
Includes: 38 h of video, 12 brief templates, community forum access. No written feedback on submitted briefs.
Everything in the self-directed tier, plus written instructor feedback on each of the 12 design briefs. Monthly live Q&A sessions are included for the duration of active study.
Response time on brief feedback: 3–5 business days. Access period: 6 months from enrolment with one-time extension available.
Pricing is available on the webinar materials page. For questions about access, write to support@adexkrost.com.
See full programme details