This page will provide information on what is covered in each lecture along with pdf of the hand written material from each class. This page will be updated as the class progresses.

The course notes is available from course page.

You can check here for a tentative order in which the topics will be covered and check here for a somewhat exhaustive reading list of papers that we plan to cover.

Date	Summary
-	HW 2013_01_18. Let f(.) be a convex function defined on domain D. Prove that D should be continuous and convex. 2013_01_18. Apply the necessary conditions of optimality for soft/hard SVM. 2013_01_22. Read Section 4.4.2(page 288 onwards) from here. Read about primal and dual problem, duality gap and KKT conditions. 2013_02_01. How to handle "no gap" case for string kernels? 2013_02_01. Write K_n(sx,t) in terms of K_n(s,t) and K'_i(s,t). 2013_02_12. Efficiently compute the kernel function K(S1,S2) between two sentences as K(S1,S2) = \sum_{T1,T2} K(T1,T2) where T1 and T2 are all possible valid parse trees of S1 and S2 respectively. 2013_02_26. Prove- (a) <f,g>=<g,f> (b) <f1+f2,g1+g2>=<f1,g1>+<f1,g2>+<f2,g1>+<f2,g2> (c) if <f,f>=0 then f=0 2013_03_01. Random walk kernel-How to avoid moving back to a node immediately? 2013_03_08. Modify CYK algorithm to find max parse tree (fixed w). 2013_03_12 - Express sequence parisng as a parse tree computaion problem.
08/01/2013	Warm up quiz and an Overview of the course Warm up quiz: You MUST submit your answers to these warm up quiz questions (even if you are unable to solve any of them) so that I can assess your background. Introduction to the course first through these tutorial slides and then through select tutorial slides here.
11/01/2013	Introduction to simple and composite features and illustrations of the same, Support Vector Machine Handwritten class notes. Please note the recommended reading before the next class at the end of the class notes.
15/01/2013	Soft margin Support Vector Machines, Optimality conditions Handwritten class notes. Please note the recommended reading before the next class at the end of the class notes.
18/01/2013	Constrained Optimization, Lagrange Multupliers Handwritten class notes. Please note the MADATORY HOMEWORK for the next class at the end of the class notes.
22/01/2013	Necessary KKT optimality conditions for SVM, duality, toward sufficient conditions Handwritten class notes. Please note the MADATORY HOMEWORK for the next class at the end of the class notes.
31/01/2013	SVM dual, Kernels to capture relations, String Kernels Handwritten class notes. Note connection to question 1 of warm up quiz
1/02/2013	Dynamic Programming in String Kernels Handwritten class notes.
05/02/2013	String Kernels: From quadratic to linear time Handwritten class notes.
08/02/2013	String & Tree Kernels Slides. Link for paper. Subsequence Kernels for Relation Extraction Link for paper.
12/02/2013	Kernels Induced by Context Free Grammars: Introduction to Convolution Kernels Handwritten class notes.
15/02/2013	Positive Definite Kernel Functions, Hilbert Spaces, etc Handwritten class notes.
26/02/2013	Discussion of midsem solutions, Reproducing Kernel Hilbert Space (ref: Chapter 2 of book "Learning with Kernels" Handwritten class notes.
01/03/2013	Graph Kernels, Rational Kernels Tutorial and corresponding slides. We discussed complexity results, the parts on subgraph kernels, path kernels and intro to walk kernels from here. Random walk graph kernels in some details. We only touched upon the fact that random walk kernels are special cases of rational kernels (singleton and identical input and output alphabet) without delving deeper into rational kernels. Handwritten class notes. Misc: A Short Tour of Kernel Methods for Graphs.
05/03/2013	Concluding discussion on Graph Kernels, logical kernel (kLog) Random walk graph kernels in some details. kLog: A Language for Logical and Relational Learning with Kernels. Slides can be found here Handwritten class notes.
08/03/2013	Concluding discussion on kLog, modeling structure on output space, StructSVM kLog: A Language for Logical and Relational Learning with Kernels. Slides can be found here Tutorial on StructSVM and corresponding slides. Handwritten class notes. Intro slides to Inductive Logic Programming. Detailed slides on Logic and Learning from last offerings.
12/03/2013	Modeling structure on output space, StructSVM Tutorial on StructSVM and corresponding slides. Context Free Grammar and Bottom-up parsing. As homework, go through slide numbers 16 to 27 and suggest how you would modify the CYK algorithm to produce the maximum weighted parse tree for a weighted context free grammar. Note connection of cutting plane algo employed in StructSVM to question 4 of warm up quiz
15/03/2013	StructSVM: From parse tree to sequences, the max-product algorithm (viterbi) Context Free Grammar and Bottom-up parsing. Discussion of solutions. Struct SVM for sequences and with Kernels. Graphical models Slides and notes. Read for viterbi/max-product Handwritten class notes. Homework: Get familiarised with submodularity: slides on submodularity.
19/03/2013	StructSVM: kernelised StructSVM, Intro to Graphical Models Struct SVM with Kernels. Graphical models Slides and notes. Handwritten class notes.
22/03/2013	Concluding intro to Graphical Models, equivalnce of conditional indepedence and factorisation, Markov Blankers, revisiting the idea of sum-product (belief propagation) and max-product Graphical models Slides and notes. Handwritten class notes (combined class notes for 19th and 22nd).
26/03/2013	Maximum Margin Markov Networks Running example for motivating MMNs. Maximum Margin Markov Networks. Handwritten class notes. Homework:
04/04/2013	Relaxed Integer Linear Program Inference, Total Unimodularity Handwritten notes for Linear Programming, Integer Linear Programming, Unimodularity and Submodularity. Note connection to the relation between Integer Linear Programming and Relaxed Integer Linear Programming from question 3 of warm up quiz Maximum Margin Markov Networks.
09/04/2013	Associative Markov Networks, Submodularity, Constrained Conditional Models Handwritten notes for Linear Programming, Integer Linear Programming, Unimodularity and Submodularity. Two more (summary) pages of handwritten notes. Maximum Margin Markov Networks. Max Margin for Hierarchies. Associative Markov Network. Associative Markov Network. Wikipedia disambiguation using min-cut. Constrained Conditional Models. Link to video lecture. Seminar on Constrained Conditional Models by Anamay. Submodularity: slides on submodularity.
10/04/2013	Dual (kernelised) of Maximum Margin Markov Networks, Markov Logic Networks, Satisfiabilty problem, DPLL and Walk Sat algorithm, MaxSat problem and MaxwalkSat algorithm for solving it, Max Margin Markov Logic Network, Integer linear formulation of the max sat problem Markov Logic Network: Introduction (slide nos 1 to 30). A more detailed tutorial. Max Margin Markov Logic networks (slide nos 17 to 22). Satisfiability problem, DPLL algorithm, Walk Sat algorithm and the maximum weighted satisfiability problem and MaxWalkSat algorithm (slide nos 52 to 65). Handwritten class notes.
12/04/2013 and 16/04/2013	Search space over interpretable features as a lattice, different partial orders corresponding to different types of kernels, least general generalisation, greatest lower bound, anti-monotonicity, apriori search algorithm, searching over graphs and lattices using dfs, greedy hill climbing, bfs, beam search, branch and bound search Sections 1.3.8, Sections 1.4.6 onwards of chapter 1, and chapter 2 of these notes. Chapter 6 of these notes. Handwritten class notes. Further reading: Submodularity: slides on submodularity. Efficient Feature Induction using Hierarchical Kernels. Introduction to Rational Kernels. Learning Languages with Rational Kernels: paper and talk.