Advanced level · Module 5 of 6 · 22 guided hours
Responsible Deep Learning
Address bias, privacy, consent, compute cost and misuse in deep learning, and document systems responsibly.
Learning objectives
- Identify and measure bias in vision and language models
- Handle images, voice and text data with consent and privacy
- Consider compute and environmental costs and document systems
Topics
- data set bias in vision and language
- performance across groups and conditions
- consent for images and voice
- privacy-preserving techniques
- compute and environmental cost
- model cards and red-teaming
Lessons
-
Lesson 1 Bias in Vision and Language Models
Recognise how training data shapes what deep models do well and badly, and measure it.
representation in training dataskin tone and lighting in imageslanguage coveragedisaggregated evaluation
-
Lesson 2 Consent and Privacy for Images, Voice and Text
Collect and use personal media responsibly, with consent, minimisation and protection.
informed consentfaces and voices as biometric datablurring and minimisationon-device processing and federated learning
-
Lesson 3 Compute Cost and Environmental Impact
Weigh the financial and environmental costs of training and running deep models.
training vs inference costenergy useefficient modelsproportionality
-
Lesson 4 Documentation and Red-Teaming
Document deep learning systems and test them deliberately for failures and misuse.
model cardintended and prohibited usesred-teamingfailure documentation
Practical activity
Audit a provided face-free image classifier and a text classifier for performance across conditions and languages, then write their model cards and a red-team report.
Expected outcomes
- Responsible design and auditing habits
- Complete documentation
Assessment
Audit report, model cards and red-team findings marked against a rubric.
Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.