ML BuzzWords demystified
summary
Covers concrete claims and tools: Google AutoML and neural architecture search can design nets automatically but cost huge compute so transfer learning (pretrained weights) often cuts training from days to hours; a Google-style fairness paper models a hidden unbiased dataset to reduce bias; GANs produce realistic faces, art (GAN-generated portrait $432,000), and data augmentation but also enable adversarial attacks like stickers that fool stop‑sign detectors; deep reinforcement learning runs Google’s datacenter cooling and reports ~40% energy savings, while defenses against adversarial examples remain unresolved.
AI-written summary · about
chapters (39)
- 0:00Intro
- 3:05Passing the Turing test
- 3:45Isaac Asimov
- 5:26Everything under control, someone's
- 6:20Choices made for you
- 10:09Devices reflecting stereotypes
- 11:23Aggression detected?
- 12:15Debiasing by post-processing
- 12:49Debiasing by removing features
- 13:58Correcting bias before training
- 17:59Model Agnostic Interpretation
- 19:03Model Interpretability?
- 21:26Can data scientist be replaced?
- 22:21Framing a project
- 22:55Building data pipelines
- 23:26Model development
- 26:07Automated Machine Learning : Azure ML
- 28:36Automated ML future
- 29:02Transfer skills and knowledge
- 30:42Norwegian Moose Detector
- 31:17Transfer Learning: replace prediction unit
- 31:40Transfer Learning: freeze other layers
- 32:00Pretrained networks
- 33:27GAN: Generate the data?
- 34:38GAN: Architecture
- 36:23GAN applications : data augmentation
- 37:31GAN : Image reconstruction
- 37:43Image to Image transition
- 38:04Al creates modern art
- 39:52GAN : Adversarial attacks
- 41:18Reward and Punishment
- 41:37Feedback from reality
- 42:49Deep Reinforcement Learning
- 43:43Alpha Go Victory
- 45:23Boston Dynamics: back to basics
- 50:02Adaptability
- 50:34Correlation is not Causation
- 51:48Cause-Effect connection
- 52:19Causal Learning
related talks
50mML Buzzwords Demystified
30mHow to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML
57mBuilding smarter solutions with Machine Learning, from magic to reality
44mBuilding smarter solutions with no expertise in machine learning
47mBuilding smarter apps with Machine Learning, from magic to reality
10mMachine Learning: The next great leap
27mKeynote - Google Cloud, Machine Learning
32mSponsored Session: Beyond the Buzz: Navigating AIs Myths and Realities