Completed
More about optimization in next talk, including recent works using trajectory analysis for gradient descent
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Mathematics of Machine Learning: An Introduction - Lecture 1
Automatically move to the next video in the Classroom when playback concludes
- 1 Date & Time: Tuesday, 12 February,
- 2 Date & Time: Tuesday, 12 February,
- 3 Date & Time: Wednesday, 13 February,
- 4 ICTS-TIFR: An Overview
- 5 ICTS and its Mandate
- 6 The ICTS Campus - Imagined 2012
- 7 The ICTS Campus - Realised 2017
- 8 ICTS Research - Structure
- 9 ICTS Programs
- 10 What ICTS is Not
- 11 ICTS Programs - Format
- 12 ICTS Programs - Duration
- 13 ICTS Programs - Organisation
- 14 ICTS Programs - Directions
- 15 ICTS Programs - Numbers
- 16 ICTS Programs - A Sampling
- 17 ICTS Outreach - Initiatives
- 18 ICTS Outreach-Kaapi with Kuriosity
- 19 Thank You See You Again at ICTS
- 20 What is machine learning and deep learning?
- 21 Machine learning ML: A new kind of science
- 22 Talk overview
- 23 Part 1 - Mathematical formalization of Machine Learning ML
- 24 Old Idea: Curve fitting Legendre, Gauss, c. 1800
- 25 Example: Learning to score reviews
- 26 Example: Learning to rate reviews contd
- 27 ML ~ finding suitable function "model" given examples of desired input/output behavior
- 28 Formal framework
- 29 Training via Gradient Descent "natural algorithm"
- 30 Subcase: deep learning* deep models = "multilayered"
- 31 Summary so far:
- 32 Unsupervised learning no human-supplied labels
- 33 A Language model baby "word2ver" [Mikolov et al'1 3]
- 34 Properties of semantic word vectors
- 35 Sequential decision-making framework
- 36 Game-playing via Deep Learning crude account of Alpha-Go Zero
- 37 Part 3 - Toward mathematical understanding of Deep Learning
- 38 Special case: deep learning deep = "multilayered"
- 39 Some key questions
- 40 Analysis of optimization
- 41 Black box analysis sketch
- 42 More about optimization in next talk, including recent works using trajectory analysis for gradient descent
- 43 Why no overfitting?
- 44 Part 4 - Taking stock, wrapping up
- 45 1. Imitation approach has not worked well in the past: airplanes, chess/go etc.
- 46 Sample Task: "Decoding" Brain fMRI [Vodrahalli et al, Neurolmage'17]
- 47 Brain regions useful for decoding
- 48 Can Machine Learning thrive in India?
- 49 Concluding thoughts on ML
- 50 Q&A
