Talk Responses will consist of a brief summary and reflection of the talk based on posted questions. You will choose one prompt and write a response. Responses should be approximately 1 page double-spaced, 1 inch margins and 12 point font. Each Talk Response will be worth 1 point and are due at 6pm Pacific time the Sunday after the talk it focuses on. Sample of a talk response

Submission: Submit the talk response directly in the text box of the Canvas assignment. Please do NOT paste a link.

Talk Responses will also include an OPTIONAL journal prompt. You do not have to turn this in; this is meant as a way for you to reflect on the talks as well as your own experiences, thoughts, and feelings.

Peer Discussions provide you with an opportunity to engage with your peers. Sharing and discussing with others is an important part of learning, so please engage thoughtfully and respectfully. Peer Discussions will be worth 1 point and are due at 6pm Pacific time one week after the Talk Response is due (with the exception of the last Peer Discussion [see schedule]).

1. Post your discussion question response under the corresponding DQ thread on Piazza. If you want to be anonymous, that is fine. Please post to Piazza AFTER the Talk Response deadline. 
2. Read two other peer’s posts and post a comment on each. Comments should be thoughtful. You may respond to a comment as well. In short, you need to post twice. This needs to be done with your name. 
3. Screenshot your two responses (include the context that lead to your comment).

Submission: Submit the two screenshots on Canvas for credit.

Dive into Deep Learning

B95

(D2L refers to Dive into Deep Learning; B95 refers to Bishop, 1995. For example, B95: 3.1-3.2 refers to Chapter 3, sections 1 and 2 in Bishop).

Linear Algebra , Probability and Calculus background in D2L, chapter 2, as well as in D2L chapter 18, Mathematics for Deep Learning. See also B95 Ch’s 1 and 2).

Week 1: Single-layer networks: Linear regression (B95: 3.4; D2L: 3.1-3.3), logistic regression (D2L: 11.3.1-11.3.2; B95: 3.1.3), and softmax regression (D2L: 3.4-3.7; B95: 3.1-3.2, 6.1, 6.7, 6.9).

Week 2: Perceptrons (B95: 3.5); Activation functions; Forward & Back propagation (B95: 4.1-4.3, 4.8; D2L: 4.1-4.3, 4.7, 11.3); Representations (see Chapter8_with_notes_from_GWC, hinton86, Sejnowskispeech1987 under Readings).

Week 3: Tricks of the trade (D2L: 11.4-11.11, see lecun_98_efficient_backprop.pdf in Readings, Batch_Normalization.pdf in Readings); objective functions (B95: 6.1, 6.7, 6.9, see also chopra05.pdf & hadsell_chopra_lecun_2006.pdf in Readings); improving generalization (D2L: 4.4-4.6). See also D2L 4.9.

Week 4: Convolutional Networks (D2L: Ch 6; also see Stanford convnets course CS231n, Chapter8_with_notes_from_GWC pp. 348-352 in Readings, lecun89e in Readings).

Week 5: More Convnets (D2L: Ch 7.)

Review session Wednesday, February 3^rd, 7-9PM (will be recorded)

Midterm Friday, February 5^th, Online. There will be a lecture also!

Week 6: Recurrent networks (D2L: Ch8, Chapter8_with_notes_from_GWC pp. 354-361, papers by Jordan, St. John, Elman under Readings); LSTM networks (D2L: Ch 9)

Week 7: Attention networks (D2L: 10.1-10.2) more applications (see DRAM and neural turing machine paper in readings (remind me to post if not there); Transformers (D2L: 10.3)

Week 8: Reinforcement Learning See lecture notes.

Week 9: More Reinforcement Learning; GANs (D2L: Ch 17)

Review session Wednesday, March 3^rd, 7-9PM (will be recorded)

Midterm , FridayMarch 5^th, Online.  cumulative (emphasis on second half)

There will be a lecture also!

Week 10: Selected recent papers; Ethics of AI