Few-Shot Defect Classifier for a Fast-Onboarding Industrial AI Vendor
Overview
What this challenge is about.
You receive a multi-customer defect dataset (8 historical customers, 4-6 defect classes each). Treat 6 customers as the meta-training set and 2 as the held-out 'new customer' scenarios. Train a prototypical network with a ResNet-18 backbone, using episode-based meta-training. Evaluate at K=5, 10, 20 shots per class on the held-out customers and compare against a fine-tuning baseline (pretrained ResNet-18, classifier head fine-tuned on the K shots only). Report per-K accuracy and time-to-first-classifier in minutes. Deliver a 3-page recommendation memo.
The Brief
What you'll do, and what you'll demonstrate.
Build a few-shot defect classifier that lifts new-customer cold-start accuracy under K=5/10/20 shots over standard fine-tuning.
Earning criteria — what you'll demonstrate
- Apply meta-learning (prototypical networks) to a real industrial cold-start problem
- Design episode-based meta-training and held-out customer evaluation
- Compare meta-learning honestly against a fine-tuning baseline
- Translate few-shot results into customer-onboarding time savings
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
ML Researcher
Prototypical-network implementations with rigorous baselines on industrial data are the ML-researcher's headline portfolio piece for any cold-start-prone AI vendor.
This challenge sharpens
- meta-learning
- few-shot-learning
- prototypical-networks
Applied AI Scientist
Translating few-shot accuracy lifts into customer-onboarding time is exactly the applied-AI-scientist's contribution to GTM-bound product roadmaps.
This challenge sharpens
- few-shot-learning
- transfer-learning
- convolutional-neural-networks
Computer Vision Engineer
Building a cold-start CV system that ships in minutes per new customer is a transferable CV-engineer skill in any visual-inspection or fast-scaling vision product.
This challenge sharpens
- convolutional-neural-networks
- transfer-learning
- pytorch