Ananta Anil Shahane

A recent masters graduate from Universiteit Leiden, with wide range of experiences actively seeking for opportunities.

Education
  • Universiteit Leiden, Master of Science
    Major: Foundations of Computing.
    Published Paper: A Corridor Model Evolutionary Algorithm, for Fast Converging Green Vehicle Routing Problem.
    Grade:7.5

    Subjects in Masters

    Quantum Algorithms, Modern Game AI, Introduction to Machine Learning, Evolutionary Algorithms, Embedded Systems, Combinatorial Algorithms, Introduction to Deep Learning, Multimedia Systems, Software Testing & Verification, Robotics, Computational Models and Semantics
  • Pune University, Bachelor of Engineering
    Major: Computer Engineering.
    Thesis: A Scalable Broker Model for Micro-Service Orchestration.
    Grade:64%

    Subjects in Bachelors

    Object Oriented Programming, Microprocessor Architecture, Operating Systems and Administration, Data Structure and Problem Solving, Computer Graphics, Digital Signal Processing, Software Engineering, Digital Communication and Wireless Sensor Network, Embedded Operating Systems.
Projects
Evolutionary Strategy
Feature Importance

I designed and implemented a sophisticated solution capable of figuring out feature importance of time series. This innovative approach effectively computes and provides an approximation of the specific weightings assigned to the individual base time series that collectively contribute to the final time series result.

Question: You are given a bunch of time series, it can be completely random and not necessarily sinosuidal waves, and a weighted sum of those time series. How do you solve for the contribution of each of the time series?
Solution: I used an Evolutionary Strategy, that searches for the optimal weightages, and provides (close to) optimal solution for the weightages, and it has been proven to return error less than 10-5 for 40 time series and perfect solution for 10 time series.
Optimisation: I implemented a solution that modulates the search parameter as a function of the error and number of generations since stagnation, giving a really fast and conveerging algorithm, often taking ~400 iterations to find optimal solution when solving for 10 time series.
Philosophy: One can visualise it as a light rain across whole domain of the rough terrain (search parameter proportional to error), then eventually it rains heavily in one of the deep trenches (exploiting local minima), and when it is filled it moves on to other trench (search parameter proportional to stagnation period).
Repository
Evolutionary Strategy
Normal Distribution
Feature Importance
Swift
Python
Hobby Project; October 2023.
Master's Thesis
A Corridor Model Evolutionary Algorithm for Fast Converging Green Vehicle Routing Problem

Remodelled genetic algorithm for green vehicle routing problem accounting for customers' locality, and engineered corridor model. Reduced average AUC for 92 benchmarks of 500 iterations by 10%.

Question: You have a bunch of customers awaiting to receive their orders, and you are given one depot and a few vehicles. Can you find a tradeoff between cheapest-lowest fuel consumption-and shortest path to serve all the customers?
Solution: I used an NSGA-II (Non-Dominant Sorting Genetic Algorithm-II), for selection process, and introduced Corridor Model from Evolutionary Strategies, into this Discrete Mathematics problem.
Result: This resulted in an algorithm that is n log(n) times slower in execution, but in return converges much faster over the generations, thanks to disciplined repair operators in the mutations.
Philosphy: A random mutation can provide one with optimal solution every once in a while, but choosing carefully early in the optimisation period leads to much more consistent convergence to optimal distance, which turns into extremely good population for exploration of optimal in other objectives.
Repository
Read Thesis
NSGA-II
Evolutionary Algorithms
Multi Objective Optimisation
Swift
Python
Leiden Institute of Advanced Computer Science, with Yingie Fan and Niki van Stein; 2022-2023.
Evolutionary Algorithm
Evolutionary Intelligence

Designed genetic algorithm for searching Deep Learning Architecture for image detection; revealed the link between adjacency matrices and the crossover operator, which improved both convergence speed and stability of the algorithm.

Question: Say you want to perform image detection using deep learning architecture, but there is no know architecture that is discovered to do so; how does one find the desired DL architecture?
Solution: I designed a Genetic Algorithm, that explores various architecture, and is quickly able to converge to an architecture that can perform image detection with 95.05% accuracy.
Optimisation: I discovered a link between the n-point crossover with n-layer architecture, that would significantly improve the performance of the GA in finding the DL Architecture more consistently and more quickly.
Philosphy: A n-point crossover is more likely to share optimal layer found by random mutation than any other crossover, as it is less likely to break a parents' layer into multiple parts, preserving the better performing layers, and potentially introducing newer better layers, improving performance.
Repository
Genetic Algorithm
Deep Learning
iOH Analyser
Python
Leiden Institute of Advanced Computer Science, with Hao Wang; Spring 2023.
Embedded Systems
Sobel on a Chip

Programmed Xilinx FPGA using the Daedalus Framework to execute edge detection using static affine programming paradigm. Used only 66% of the instructed channels in KPN, earning recognition as the top performing software among peers of about 3.75 FPS.

Question: Given a Xilinx FPGA, the goal was to program it to be able to do sobel (edge detection) at as high frame rate as possible.
Process: While working on the assignment, I noticed that the SobelX has a 0 row and sobelY has a 0 column. Hence during the design phase of the Kahn Process Network, I decided to ignore the associated channels, reducing the number of channels required down to 66%.
Results: The expected performance was 2.8-3.5 FPS, but my algorithm provided 3.75 FPS, earning the recognition.
The repostory is not available, due the data being on server, which was not downloaded.
Repository
C
Deadalus Framework
Leiden Institute of Advanced Computer Science, with T.P. Stefanov; Spring 2023.
Quantum Algorithms
Randomised Measurement Toolbox

I researched classical shadows in quantum computing and implemented a randomized measurement toolkit, validated its performance, and utilized it to efficiently evaluate the matrix of underlying quantum circuits. This work, detailed in the research paper here, contributed to reducing the number of required samples for accurate quantum circuit analysis, and I implemented the algorithm and validated the results it claimed.

Quantum Algorithms
Linear Algebra
Python
cirq
Openfermion
Leiden Institute of Advanced Computer Science, with Vedran Dunjko; Spring 2022.
Modern Game Artificial Intelligence Applications
2-D Search and Replace Algorithm & Minecraft City Builder

In Modern Game Artificial Intelligence Applications, I worked on building a 2-D Search and Replace Algorithm, that is an iteration on the procedural content generation allowing one to build 2-D map step by step using regex like commands. I was also involved in building a home using PSG in a minecraft map, using http interface to do it.

Report on 2-D Search and Replace
Minecraft Procedural Content Generation
Python
gdmc
regex
Procedural Content Generation
Leiden Institute of Advanced Computer Science, with Mike Preuss & Piere Lulof; Spring 2022.
iOS / macOS Development
Today Productivity

I conceived, developed, published, and provided ongoing support for a cross-platform application called Today Productivity. This multifaceted app, available on iOS, iPadOS, and macOS, was meticulously designed to empower users in planning and executing their daily tasks. It offered a comprehensive feature set, including support for Widgets and seamless integration with iCloud for shared and persistent storage across all of the user's devices.

More Info
Xcode
RDBMS
SwiftUI
CloudKit
Widgets
AppStore Publishing
Web Dev.
Hobby Project; 2019-2023.
Robotics
SLAM

Utilised time averaged K-Nearest Neighbours Algorithm to approximate discrimination between walls and furniture. Helped ease the navigation of the Pioneer P3DX two-wheeled robot with SICK TiM310 LIDAR sensor to different rooms in a home settings.

Watch Video
Python
Machine Learning
SLAM
CoppeliaSim
Trignometry
Leiden Institute of Advanced Computer Science, with E.M. Bakker; Spring 2022.
Bachelors Thesis
Gilmour Proxy

Built a proxy for a micro-service broker system named Gilmour. The broker runs only on Golang and ruby. Gilmour proxy makes the broker system accessible for all other languages by using https protocol. This broker can handle thousands of micro-services written in different programming languages. Responsibilities include developing and debugging the proxy. Documentation using LateX and markdown. Containerisation and monitoring with Docker and Prometheus.

Repository
Golang
Docker
Prometheus
Redis
Micro-Service
Oogway Consulting, with Aditya Godbole; 2017-2018.
Audio Steganography
StealthAudio

Designed and build a steganography algorithm, that takes a text and an audio file *.wav from the user, and hides the text into the provided file, it also able to pull out the text embedded in the audio file. The audio file sounds un tampered despite having hidden message in it.

StealthAudio
C++
qt
File System
bitstream
ASCII Code
Pune University, Mini Project; 2017.
Miscellaneous
Category Details
Programming Languages I am proficient in Swift a protocol oriented programming language, C++ an object oriented programming language, comfortable with Python and familiar with Golang. I can also do scripting like writing daemon files, and running processes using bash.
Frameworks Over the years I have accquired experience with a lot of different frameworks, and have gotten stronger at referring to documentation to get started quickly. These framweorks include Scikit Learn, Cirq, Openfermion, Json, Keras, gdmc, SwiftUI, CloudKit, Qt and many more.
Operating Systems I mostly prefer the stability of Unix and Linux based operating systems, so I have experience in using macOS and Linux Fedora Distro, as a power user, and some familiarity with Windows. I am fluent in using the macOS and Linux terminal and have accquired RHEL Certification for it.
Certification Course edX 2019, Fundamentals of Red Hat Enterprise Linux
Awarded for 100% completion of course, with 8 assignments, 6 weeks focusing on Linux Administration.
Certification Course University of California Invine, Coursera 2017
Awarded for 98% grade of course, spanning 4 months and 16 assignments, focusing on various frameworks available on iOS platform.
Awards
  • 91st percentile in National Cyber Olympiad, India in 2010.
  • 1st among 120 in Junior College Level, Physics.
  • Indian Association of Physics Teachers 1st of 23 in 2013-2014.
Conferences Attended
(2017-2023)
  • try! Swift Banglore (2017).
  • Kernel Meetups 2017-2018
  • Free Open Source Software Meet (2018)
  • GopherCon (2018)
  • Google Meetup (2018)
  • Quantum Games Leiden (2023)
  • Quantum Amsterdam
  • Joint Lecture on Evolutionary Algorithms (JoLEA)-CWI
  • GECCO 2024, Melborne Australia*.
Activities
Relax~
  • Biking 1800+km since 2021.
  • Placed 17th in National DOJO Tournament India for Tekken 7.
  • Built a PC with latest most powerful hardware for available budget.
  • Video editing using Final Cut Pro for YouTube Channel-now depricated.
  • High altitude trekking (3.8 km above sea level), to Pindari Glacier in upper Kumamon Himalayas.
Languages
I speak
  • Fluent English.
  • Fluent in Hindi.
  • Fluent Marathi.
  • Basic Dutch.