About Me

Mohammad Yousefi

AI Researcher

About Me

My name is Mohammad. I am an AI developer/researcher, currently pursuing a doctorate degree at the Australian National University. My research revolves around planning under uncertainty. In particular, Fully Observable Non-Deterministic (FOND) Hierarchical Task Network (HTN) planning, which aims to develop intelligent agents capable of systematically face unpredictable environments. What draws me to this field is the parallel between planning under uncertainty and the beautiful chaos of the real world. The fact that us, humans, are relentless non-deterministic planners.

When I'm not immersed in the world of AI research, I enjoy exploring the creative possibilities of game development. Although time constraints often limit my ability to fully engage in the development process, my fascination with crafting interactive experiences remains strong. What I find particularly appealing about game development, as opposed to AI, is the freedom to create worlds that aren't necessarily bound by human-centered design. In games, we can alter the laws of physics, create unique visual aesthetics, and design environments that defy reality. This level of creative control allows for a kind of magic that AI, in its current form, cannot easily replicate. I believe that games represent the highest form of art, seamlessly blending elements of painting, poetry, and music into an interactive medium. Some of my favorite titles, such as Detroit: Become Human, The Stanley Parable, and Life is Strange, exemplify the power of games to tell compelling stories, evoke deep emotions, and challenge our perceptions.

I'm a bookworm through and through! Give me anything from a rigorous non-fiction to a mind-bending fiction, and I'm in my happy place. My love for reading has not only enriched my personal life but also significantly influenced my academic pursuits. In fact, it was Douglas Hofstadter's "Gödel, Escher, Bach: An Eternal Golden Braid," that sparked my fascination with the intricate connections between mathematics, art, and philosophy; ultimately inspiring me to embark on my Ph.D. journey. Beyond this, I have a deep appreciation for Russian literature, finding myself captivated by the works of literary giants such as Dostoevsky, Tolstoy, and Chekhov. Their profound insights into the human condition and their masterful storytelling continue to inspire and shape my worldview.

Education

Degree: Doctor of Philosophy
Field: Engineering and Computer Science
Institute: Australian National University
Location: ACT, Australia
Timeline: Aug 2023 - Present
Thesis: FOND HTN Planning
Supervisors: Dr. Pascal Bercher
Dr. Patrik Haslum
Dr. Ron Alford

Degree: Master’s Degree
Field: Computer Engineering - Artificial Intelligence
Institute: Kharazmi University
Location: Tehran, Iran
Timeline: Sep 2019 - Sep 2022
Thesis: A Hybrid RL-MIP Multi Objective Routing Algorithm
Supervisors: Dr. Farshad Eshghi
Dr. Manoochehr Kelarestaghi

Degree: Bachelor’s Degree
Field: Computer Engineering - Software
Institute: Razi University
Location: Kermanshah, Iran
Timeline: Sep 2014 - Sep 2019
Project: Implementation of a Cryptocurrency in Python
Supervisors: Dr. Hamed Monkaresi

Work Experience

Role: Tutor
Employer: Australian National University
Location: ACT, Australia
Timeline: Feb - May 2024, and Jun - Oct 2024
Description: As a tutor of the Computing Team Project course, also known as TechLauncher, I have been responsible for providing guidance to 8 teams consisting of more than 50 students, primarily Master of Computing, as they work on various industrial projects ranging from developing e-commerce websites to cutting-edge computer vision applications for two semesters. The focus was on applying software development methodologies such as Agile to real-world projects. This course helps students navigate the complexities of industrial projects, develop essential skills in teamwork, understand stakeholder's vision, and ultimately enable them to successfully deliver their projects to production.

Role: Algorithm Developer
Employer: Communere Ltd.
Location: Surrey, England (Remote)
Timeline: Apr 2022 - Jun 2023
Description: Through my engagement with Communere Ltd., a software consulting company, I was assigned to work with their client, Basemap Ltd., a provider of digital mapping and multi-modal time travel analysis solutions. My primary responsibility was to develop an advanced algorithm for public transportation planning, with a specific focus on analyzing the accessibility of critical locations such as hospitals within the public transport network. The algorithm needed to be designed from scratch to work efficiently in parallel and operate within strict memory and time limitations, as it would be running on office computers. The result has been integrated to Basemap's TRACC software, and is currently in operation.

Role: Data Scientist
Employer: Daneshgar Technology Co. Ltd.
Location: Tehran, Iran
Timeline: Jan 2021 - Jan 2022
Description: At Daneshgar Technology Co. Ltd., I was part of a team tasked with developing a distributed machine-learning algorithm for detecting anomalies in financial transactions using unsupervised learning techniques for their client, the Central Bank of Iran. The primary objective was to create a scalable solution that could process vast amounts of data efficiently while providing meaningful insights into potential fraudulent activities. The most significant challenge faced during this project was the enormous size of the dataset, which made it impossible to even load onto a single machine. To overcome this obstacle, I played a key role in transforming the raw SQL data into an extensive network of transactions and engineering novel feature embeddings based on behavioral profiles, while incorporating path analysis algorithms to detect local networks of abnormal users, enabling the identification of complex patterns and relationships within the data.

Role: Algorithm Developer
Employer: Walk4Less
Location: Tehran, Iran
Timeline: Mar 2021 - Mar 2022
Description: In a project aimed at reducing travel costs for ride-sharing platforms by optimizing pick-up and drop-off locations based on real-time traffic data, I contributed to the development of the search algorithm. The project faced challenges due to the lack of a well-established mathematical framework for Mobility as a Service optimization with uncertain endpoints. To address this, I participated in the creation of a new mathematical fomulation that captured the interplay between transportation networks and user preferences. Existing solutions lacked the ability to provide real-time optimizations, a critical requirement for the startup's success. As part of the team, I helped design and implement an advanced search algorithm that efficiently approximated optimal change in real-time, considering factors such as traffic congestion, road conditions, and user preferences.

Open-Source Projects

Title: Koala Planning Engine
Description: Koala is a model-based hierarchical planner designed to operate in non-deterministic environments. At its core, Koala implements the AO* search algorithm paired with a cascade of custom heuristics. The planner integrates a modified parser and the grounder derived from the PANDA planning system, enabling it to process lifted domain descriptions with non-deterministic effects through an extended HDDL input format. As the first planner to achieve optimal solutions for non-deterministic hierarchical planning problems, Koala fills a notable gap in automated planning technology.
Link: Github

Title: HDDL Parser
Description: HDDL Parser is a recursive descent parser designed from ground up to validate HDDL files. This tool can detect a wide range of errors ranging from simple syntactic deviations from the grammar to semantic type checking, and suggest fixes for them.
Link: Github

Publications

Title: Laying the Foundations for Solving FOND HTN Problems: Grounding, Search, Heuristics (and Benchmark Problems)
Authors: Mohammad Yousefi
Pascal Bercher
Venue: Proceedings of the 33rd International Joint Conference on Artificial Intelligence
Publisher: IJCAI
Year: 2024
Paper: Click Here to Download
Abstract: Building upon recent advancements in formalising Fully Observable Non-Deterministic (FOND) Hierarchical Task Network (HTN) planning, we present the first approach to find strong solutions for HTN problems with uncertainty in action outcomes. We present a search algorithm, along with a compilation that relaxes a FOND HTN problem to a deterministic one. This allows the utilisation of existing grounders and heuristics from the deterministic HTN planning literature.

Title: A Visual Studio Code Extension for Automatically Repairing Planning Domains
Authors: Songtuan Lin
Mohammad Yousefi
Pascal Bercher
Venue: Demonstration Track of the 34th International Conference on Automated Planning and Scheduling
Year: 2024
Paper: Click Here to Download
Abstract: We demonstrate a Visual Studio Code extension which aims at providing modeling assistance for modeling planning domains in PDDL. More specifically, the extension can identify potential flaws in a domain and propose respective corrections by taking as input a set of counter-example plans, which are known to be valid but actually contradict the domain. Those input plans shall be provided by the user. The flaws are then identified and corrected by making changes to the domain so as to turn those plans into solutions, i.e., the changes are regarded as potential corrections to the domain. Currently, the extension only supports corrections that add predicates to or remove predicates from actions' preconditions and effects.

Academic Roles

Role: Integrated AI network HDR Group Leader
Venue: Australian National University
Timeline: Dec 2024 - Present

Role: Program Committee Member
Venue: Hierarchical Planning (HPlan) Workshop
Year: 2024

Role: Subreviewer
Venue: European Conference on Artificial Intelligence (ECAI)
Year: Omitted for Anonymity

Role: Reviewer
Venue: Human-Aware and Explainable Planning (HAXP) Workshop
Year: Omitted for Anonymity

Role: Elected Member
Venue: Students’ Scientific Association, ECE. Department, Razi University
Timeline: Sep 2015 – Sep 2016

Certificates

Course: Graph Analytics for Big Data
Issue Date: Oct 2021
Authorizer: UC San Diego
Issuer: Coursera
Link: Click Here to Verify

Course: Number Theory and Cryptography
Issue Date: Aug 2021
Authorizer: UC San Diego
Issuer: Coursera
Link: Click Here to Verify

Course: Fundamentals of Scalable Data Science
Issue Date: Apr 2021
Authorizer: IBM
Issuer: Coursera
Link: Click Here to Verify

Course: Anomaly Detection in Time Series Data with Keras
Issue Date: Jan 2021
Authorizer: Coursera
Issuer: Coursera
Link: Click Here to Verify

Course: Finance & Quantitative Modeling for Analysts Specialization (4 Courses)
Issue Date: Dec 2020
Authorizer: Wharton School, University of Pennsylvania (Online)
Issuer: Coursera
Link: Click Here to Verify

Course: Sequences, Time Series and Prediction
Issue Date: Oct 2020
Authorizer: DeepLearning.AI
Issuer: Coursera
Link: Click Here to Verify

Skills

Languages: Rust, Python, C#, C/C++, HTML, CSS
Machine
Learning:
Pandas, PyTorch, Keras, Scikit-Learn, PyG, Reinforcement Learning
Databases: Postgre, Redis
Distributed
Computing:
Apache Spark
Misc.: Git, Linux, Flask