Self-Supervised Pretraining for a Pathology Foundation Vendor
Overview
What this challenge is about.
You receive a public pathology dataset (about 80,000 unlabeled whole-slide-image patches plus a labeled 8,000-patch subtype-classification subset across 4 classes). Pretrain a ResNet-50 with DINO on the unlabeled patches, then fine-tune on the labeled subset and compare to (a) ImageNet-pretrained ResNet-50 fine-tuned identically, (b) random-init ResNet-50 fine-tuned identically. Report macro-F1 + per-class F1, plus an honest discussion of compute cost. Deliver a 4-page memo.
The Brief
What you'll do, and what you'll demonstrate.
Determine whether self-supervised pretraining on unlabeled pathology patches usefully beats ImageNet pretraining for downstream subtype classification.
Earning criteria — what you'll demonstrate
- Apply self-supervised pretraining (DINO) to a real medical-imaging domain
- Compare ImageNet vs. self-supervised vs. random-init transfer fairly
- Quantify compute cost alongside accuracy gain
- Recommend a pretraining strategy with explicit compute trade-off
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
Pretraining-strategy studies on real medical-imaging data with honest compute reporting are the ML-researcher's headline portfolio piece at pathology-AI startups.
This challenge sharpens
- self-supervised-learning
- transfer-learning
- medical-imaging
Computer Vision Engineer
Building reusable pretrained backbones is core CV-engineer territory at any foundation-model-bound healthtech team.
This challenge sharpens
- convolutional-neural-networks
- self-supervised-learning
- pytorch
Applied AI Scientist
Translating pretraining gains into a quarterly compute-budget recommendation is the applied-AI-scientist's contribution at any AI-forward biotech or healthtech.
This challenge sharpens
- transfer-learning
- self-supervised-learning
- model-evaluation