- 46
- 3 507
Krunal Patel
Canada
เข้าร่วมเมื่อ 4 เม.ย. 2009
PhD in Applied Mathematics (Operations research). Software Engineer at Arista Networks.
I graduated from BITS Pilani Goa Campus in 2015 with B.E. (Hons.) Computer Science and M.Sc. (Hons.) Mathematics. After graduation, I worked with Google from 2015 to 2020. Left Google in 2020 to start my PhD under the supervision of Prof. Andrea Lodi and Prof. Guy Desaulniers.
My research interests:
Discrete optimization
Column Generation
Machine Learning.
My non-academic interests:
Chess
Work life balance
Study skills
Productivity
Buy me a coffee: www.buymeacoffee.com/krooonal
I graduated from BITS Pilani Goa Campus in 2015 with B.E. (Hons.) Computer Science and M.Sc. (Hons.) Mathematics. After graduation, I worked with Google from 2015 to 2020. Left Google in 2020 to start my PhD under the supervision of Prof. Andrea Lodi and Prof. Guy Desaulniers.
My research interests:
Discrete optimization
Column Generation
Machine Learning.
My non-academic interests:
Chess
Work life balance
Study skills
Productivity
Buy me a coffee: www.buymeacoffee.com/krooonal
ML Augmented Heuristics
In this episode of the Reading Club, we explore a part of the research paper "Machine Learning Augmented Branch and Bound for Mixed Integer Linear Programming" by Lara Scavuzzo and others. This paper delves into how machine learning can enhance various components of solving MILPs.
In this video, I focus on ML-augmented primal heuristics.
Original paper: link.springer.com/article/10.1007/s10107-024-02130-y
Timestamps
0:00 Intro
1:04 Predicting Solutions
8:38 Predicting Neighborhoods
12:39 Predicting Schedules for exising heuristics
18:46 Summary
#MIP #MachineLearning #Optimization #PrimalHeuristics #OperationsResearch
Buy me a coffee: www.buymeacoffee.com/krooonal
In this video, I focus on ML-augmented primal heuristics.
Original paper: link.springer.com/article/10.1007/s10107-024-02130-y
Timestamps
0:00 Intro
1:04 Predicting Solutions
8:38 Predicting Neighborhoods
12:39 Predicting Schedules for exising heuristics
18:46 Summary
#MIP #MachineLearning #Optimization #PrimalHeuristics #OperationsResearch
Buy me a coffee: www.buymeacoffee.com/krooonal
มุมมอง: 35
วีดีโอ
Complete Branch and Price implementation for cutting stock
มุมมอง 84หลายเดือนก่อน
In this video I implement branch and price (with the help of AI code generators) for the cutting stock problem and discuss major challenges involved in it. My PhD project CG Framework can be found at github.com/krooonal/col_gen_estimator/tree/master/col_gen_estimator Timestamps 00:00 Intro 00:33 Cutting Stock problem 05:21 Branch and Price implementation 37:12 Further improvements Buy me a coff...
How Column Generation helps with better linear relaxation
มุมมอง 862 หลายเดือนก่อน
In this video we go through the example of bin packing and understand why CG model gets better relaxation using Dantzig Wolfe decomposition theory. Timestamps 0:00 Intro 0:35 Bin Packing 8:59 CG model 21:59 Dantzig Wolfe Decomposition 29:52 Better relaxation 33:49 Bin Packing revisited Buy me a coffee: www.buymeacoffee.com/krooonal
MPT Ch 7 Calculating Efficient frontier using single index model
มุมมอง 515 หลายเดือนก่อน
We learned how to calculate efficient frontier in chapter 4. In this chapter we learn how to calculate efficient frontier for single index model and constant correlation model. Buy me a coffee: www.buymeacoffee.com/krooonal
Discounted Pseudocost in MILP
มุมมอง 355 หลายเดือนก่อน
The project I worked on when I visited ZIB this year. I also presented this work at ISMP 2024 in Montreal. Arxiv link: arxiv.org/abs/2407.06237 Editor: Saad Siddiqui Thumbnail design: Marouane FELLOUSSI Buy me a coffee: www.buymeacoffee.com/krooonal
Building a CP style Sudoku solver
มุมมอง 658 หลายเดือนก่อน
I found this question on LeetCode about building a sudoku solver. I decided to build a constraint programming style sudoku solver (with branching and propagation). I describe the design and implementation process in this video. Also, it is faster than typical brute force based solutions as expected. Question link: leetcode.com/problems/sudoku-solver/description/ My code link: leetcode.com/probl...
Column Generation in Machine Learning | Krunal Thesis Defense
มุมมอง 938 หลายเดือนก่อน
This is the recording of one of my rehearsals. The background for the projects is covered in this video: th-cam.com/video/UrMqa2QNwO0/w-d-xo.html Timestamps 0:00 Disclaimer 0:08 Introduction 1:15 NOTAM project 13:40 Decision Tree project 27:01 Framework project 32:14 Reoptimization competition work 37:59 Conclusion Buy me a coffee: www.buymeacoffee.com/krooonal
Column Generation in Machine Learning| Krunal Thesis background
มุมมอง 749 หลายเดือนก่อน
My PhD is about to end. My thesis defense is on 30 April at 1 PM Eastern Time. If you want to attend the live thesis defense, please fill out this form: docs.google.com/forms/d/e/1FAIpQLSc-pXL9TwVgZWD8tHr6wqadNXVw8ZxUtmwRsedIu3LwbeOdTg/viewform?usp=sf_link Timestamps 0:00 Intro 0:19 Boolean Decision Rules 12:53 Column Generation overview 19:40 Decision Trees 23:06 My projects overview See you s...
How to win MIP Comp Competitions?
มุมมอง 699 หลายเดือนก่อน
In this video, I share some of the tips that will help you do well in the MIP computational competition. These tips are based on my personal experience as a participant (winner) and a member of organizing committee. Timestamps 0:00 Intro 0:38 How to find good ideas? 1:15 Aim for multiple ideas 1:35 Get code reviewers 2:25 Test and benchmark frequently 3:34 How to prioritize? 4:20 Go beyond comp...
MPT Ch 6 Multi index model
มุมมอง 1259 หลายเดือนก่อน
Extension of single index models. Buy me a coffee: www.buymeacoffee.com/krooonal
Dual degenerarcy in MIP
มุมมอง 2910 หลายเดือนก่อน
In this episide of the reading club, we discuss dual degeneracy in MIP. How frequently it appears? Does it increase or decrease with branching? and can we use it to solve the instance faster? Timestamps 0:00 Intro 0:48 What is dual degeneracy? 5:30 Why care about it? 9:43 Degeneracy measures 14:37 Why use VC ratio? 15:37 How degeneracy changes with branching? 19:41 Cloud Intervals 27:21 Conclus...
MPT Ch5 Single index model
มุมมอง 4010 หลายเดือนก่อน
Single index model to simplify the inputs. Buy me a coffee: www.buymeacoffee.com/krooonal
Pseudo costs Deep Dive
มุมมอง 3810 หลายเดือนก่อน
In this episode of reading club, we discuss the implementation details of pseudo costs from SCIP's code. We will learn some details that are usually not taught in the classroom. Timestamps 0:00 Intro 0:54 Basics of Pseudo costs 5:16 Why learn pseudo costs? 7:55 Pseudo costs in CP-SAT solver 10:22 When LP is not solved 16:35 Valid pseudo cost updates 21:25 Summary 26:26 Hulk's message Buy me a c...
MPT Ch4 Techniques for calculating EF
มุมมอง 5311 หลายเดือนก่อน
Modern Portfolio Theory In this episode, we talk about techniques to calculate the efficient frontier. Timestamps 0:00 Introduction 2:45 Short sales RF lending 13:31 Example 15:33 Short sales No RF lending 20:27 No Short sales RF lending 24:03 No Short sales No RF lending 26:24 Additional constraints 29:48 Conclusion 31:39 Hulk's message Buy me a coffee: www.buymeacoffee.com/krooonal
Conflict analysis ft Pseudo Boolean reasoning
มุมมอง 28ปีที่แล้ว
In this episode of reading club, we discuss the paper "Improving conflict analysis in MIP solvers by Pseudo Boolean Reasoning" by Gioni, Timo, Ambros, and Jakob. Previous video on infeasibility analysis: th-cam.com/video/YWna9iU8w5Y/w-d-xo.html Timestamps 0:00 Intro 0:25 Conflict propagation in SAT 9:46 Slack 13:35 Why reduce (example) 18:54 Reduce 21:14 Saturation 24:11 Division 27:14 MIR Cuts...
MPT Ch1 Opportunity set and indifference curves
มุมมอง 292ปีที่แล้ว
MPT Ch1 Opportunity set and indifference curves
Adaptive large neighborhood search (ALNS)
มุมมอง 499ปีที่แล้ว
Adaptive large neighborhood search (ALNS)
Symmetry in ILP [P4]: Pruning search tree
มุมมอง 16ปีที่แล้ว
Symmetry in ILP [P4]: Pruning search tree
Symmetry in ILP [P3]: Symmetry breaking inequalities
มุมมอง 36ปีที่แล้ว
Symmetry in ILP [P3]: Symmetry breaking inequalities
Symmetry in ILP [P2]: The partitioning problem
มุมมอง 52ปีที่แล้ว
Symmetry in ILP [P2]: The partitioning problem
MIP Presolve [Part 9]: Node presolve and summary
มุมมอง 19ปีที่แล้ว
MIP Presolve [Part 9]: Node presolve and summary
MIP Presolve [Part 8]: Complementary slackness in MIP
มุมมอง 45ปีที่แล้ว
MIP Presolve [Part 8]: Complementary slackness in MIP
This is fascinating work! I'm excited to have discovered this channel.
Thank you so much for the video 🌷, but I wonder about Sonnerat's work in predicting neighborhoods.
I haven't looked at that paper in detail. But from what I understand, they find the best solution explicitly (with some unfixing restrictions), and then use that solution to learn which variables should be unfixed for the given solution.
@@krooonal Thank you
This is exciting! And I'd love to implement your chapters like the efficient frontier. I've used ceres solver for MILP. Is there a simpler stack to boilerplate data, graphing and perhaps market execution?
Modern profolio theory is outdated.
Ok.
Hi Krunal, I saw your courses on Codedamn, its good course but what you have taught there is not enough for data science and machine learning. can you please make more courses on Math for Data Science and ML please...
I will try my best to find time for recording more courses.
Krunal, excellent work . The examples at the beginning of the video were self explanatory. Happy to know that CG has wide applications in ML too . Keeping posting such good stuff . Will try to attend your thesis as well
How to start with cpp 😢 Sir we need tips We are beginners Please sir
If by CPP you mean constraint programming problems, then taking up discrete optimization modeling courses on Coursera is a good start. If you are referring to C++ programming language, I have no recommendations. This video series is not about programming languages.
Hi Krunal, what’s the tablet you use for your videos ?
A cheap one! They are not sponsoring the videos, so not going to reveal the brand name.
new suscriber!! tysm keep doing this❣