Deep Learning
Milestone 2 of 4 · Deep Learning season · The Applied AI team

The run that wouldn't converge

The first training run for the referral-form model fails: the loss spikes, turns into NaN, and the model gives different answers for the same form. Write the optimisers and normalisation layers yourself, read the curves, and fix the training script. Includes 4 hands-on Python coding tasks (NumPy), checked by automated tests.

Rigour50Communication50Ethics50Impact50Teamwork50
Mon 9:30The run that wouldn't convergeMon 13:30Code it: momentum and Adam from scratchTue 10:00Code it: batch norm and dropoutWed 11:00Code it: warm-up, cosine decay and gradient clippingWed 15:00Code it: fix the broken training scriptThu 10:00Read the curvesThu 16:00Thursday review
Mon 9:30The run that wouldn't converge#applied-ai
Huddle# applied-ai# general# data-eng# randomDirect messages● Maya● Tomás● Sam● Leah● Grace
# applied-aireferral-form model
Sam Patel is typing...

Bluegum Health, Riverside Clinics and every person here are fictional. The decisions, tools and practices are modelled on how real ML teams work.

Ready