EECS Project Courses

User Tools

Site Tools

You are here: Pre-Enrollment Website for EECS 4070, 4080, 4088, 4090, 4480 » 2024-25 » Proposed Projects for the F24 Term » F24 Project Listings
Trace:
2024-25:fall:f24-listing

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
2024-25:fall:f24-listing [2024/08/27 22:42] – [LLM Native Operating Systems] hkh2024-25:fall:f24-listing [2025/01/13 16:48] (current) – lesperan
Line 10: Line 10:
  
 /** DO NOT EDIT ABOVE THIS LINE PLEASE **/ /** DO NOT EDIT ABOVE THIS LINE PLEASE **/
 +
 ---- ----
 +==== Automotive Knowledge Graph Construction and LLM Integration ====
  
 +**[added 2024-09-03]**
 +
 +**Course:**  EECS4080
 +
 +**Supervisor:**  Aijun An
 +
 +**Supervisor's email address:**  aan@yorku.ca
 +
 +**Project Description:** 
 +This project focuses on the construction and application of knowledge graphs in the automotive domain, combining natural language processing, information extraction, and large language models (LLMs). Students will explore recent advances in the field of automatic knowledge graph construction and evaluate its impact on enhancing LLM performance.
 +
 +The project involves the following steps:
 +1. Conduct a literature review on recent advances in automatic knowledge graph construction.
 +2. Select and implement an existing knowledge graph construction method, making use of available open-source code.
 +3. Create an automotive knowledge graph by applying the chosen method to one or multiple car owner's manuals.
 +4. Evaluate the accuracy of answers of an LLM with and without the constructed knowledge graph.
 +
 +**Required skills or prerequisite courses:**  
 +  - Good programming skills in Python
 +  - Familiarity with natural language processing concepts
 +  - Basic understanding of graph theory and data structures
 +
 +**Recommended skills or prerequisite courses:**
 +  - Experience with machine learning libraries (PyTorch)
 +  - Familiarity with knowledge representation and ontologies
 +  - Basic understanding of large language models and their applications
 +
 +**Instructions:**
 +Send your transcript and a short statement of motivation to [[mailto:aan@yorku.ca|Aijun An]].
 ==== Optimizing Regulatory Document Summarization for Automated Compliance Analysis in Healthcare** ==== ==== Optimizing Regulatory Document Summarization for Automated Compliance Analysis in Healthcare** ====
  
Line 413: Line 444:
  
 ---- ----
-==== TEMPLATE ENTRY 9  - PUT PROJECT TITLE HERE====+==== Mapping Stong Pond ==== 
 + 
 +**[added 2024-08-29]** 
 + 
 +**Course:**  { EECS4080} 
 + 
 +**Supervisor:**  Michael Jenkin 
 + 
 +**Supervisor's email address:**  jenkin@yorku.ca 
 + 
 +**Project Description:**  
 +Much of the surface of the planet is covered by water. Mapping and performing other tasks on these environments can be augmented through the deployment of unmanned surface vessels (USV) that can perform these tasks autonomously. This project involves using a simulator for one of the USVs in the lab to develop a strategy to perform measurements over real and simulated bodies of water (Stong Pond). 
 + 
 +**Required skills or prerequisites:**   
 +  - Ability to work independently and in groups 
 +  - Good python programming skills 
 +  - Knowledge of/interest in ROS2 would be helpful 
 + 
 +**Recommended skills or prerequisites:** 
 +  - None beyond 4080 prerequisites 
 + 
 +**Instructions:** 
 +Contact Michael Jenkin as jenkin@yorku.ca if interested 
 + 
 + 
 +---- 
 +==== Enhanced avatar for human-robot interaction ==== 
 + 
 +**[added 2024-08-29]** 
 + 
 +**Course:**  { EECS4080 } 
 + 
 +**Supervisor:**   Michael Jenkin 
 + 
 +**Supervisor's email address:**  jenkin@yorku.ca 
 + 
 +**Project Description:**  
 +Avatars have been proposed as a key element in user interface designs since the development of Microsoft's Clippy, if not before. In the lab we have been developing a Unity-based avatar that operates as the front end of a LLM-based avatar that can be deployed in various environments. This forward facing avatar provides a natural interaction with individuals in the environment, providing audio-based input and output and literally putting a face on the underlying system.  The basic goal of the project is to take the operational system and to enhance it in a number of ways, perhaps most critically through the addition of canned animation scripts that can be used by the avatar to provide a natural interaction and non-interaction appearance to the avatar. 
 + 
 +**Required skills or prerequisites:**   
 +  - Ability to work independently and as part of a team. 
 +  - Knowledge/interest in Unity and C# programming 
 +  - Ability to work with external partners. 
 + 
 +**Recommended skills or prerequisites:** 
 +  - none beyond those required for 4080 
 + 
 +**Instructions:** 
 +Contact Michael Jenkin by email (jenkin@yorku.ca) if interested. 
 + 
 + 
 +---- 
 + 
 + 
 + 
 + 
 + 
 + 
 +==== Evaluating Planning Domain Validation Tools ==== 
 + 
 +**[added 2024-09-18]** 
 + 
 +**Course:**  { EECS4080 } 
 + 
 +**Supervisor:**  Yves Lesperance 
 + 
 +**Supervisor's email address:**  lesperan@eecs.yorku.ca 
 + 
 +**Project Description:**  
 +In this project, the student will evaluate software tools (such as Val, FastDownward, and the "unquestionable parser for PDDL 3.1") that are used to validate planning domains and planning problems specified in the Planning Domain Description Language (PDDL), focussing on the STRIPS fragment.  This is part of a larger project to use Large Language Models to generate abstract planning domain models to support more efficient planning and provide explanations at an abstract level by supressing uninteresting details from domain models.  The validation tool selected will become part of a neuro-symbolic systems to help users develop such abstract planning models. 
 + 
 +**Required skills or prerequisites:**   
 +  - EECS 3401 
 +  - Python 
 + 
 +**Recommended skills or prerequisites:** 
 +  - basic knowledge of AI planning techniques and PDDL 
 +  - some experience using large language models 
 + 
 +**Instructions:** 
 +Contact instructor by email 
 + 
 +---- 
 + 
 + 
 + 
 + 
 +==== Exploring the Use of Large Language Models to Generate Control Knowledge for HTN Planning ==== 
 + 
 +**[added 2025-01-13]** 
 + 
 +**Course:**  { EECS4080 } 
 + 
 +**Supervisor:**  Yves Lesperance 
 + 
 +**Supervisor's email address:**  lesperan@eecs.yorku.ca 
 + 
 +**Project Description:**  
 +Hierarchical Task Network (HTN) planning is a popular approach to automated planning with domain-specific search control knowledge.  In HTN planning, the objective is to find a sequence of actions to perform an abstract goal task, and the plan is generated by decomposing the root goal task into simpler tasks and actions using a handcrafted set of “refinement” methods.  However, developing such a set of methods and the associated subtasks requires knowledge of the problem domain and normally necessitates the involvement of a knowledge engineer and/or domain experts. 
 + 
 +Given that Large Language Models (LLMs) can display broad world knowledge, it seems reasonable that they should be able to help generate methods and tasks for a given HTN planning domain and root goal task.  That is, given the domain’s atomic tasks/operators, the root goal task, and the initial state specification, the LLM should be able to fill in the methods and abstract tasks (possibly with some human help), in order to generate a complete HTN planning problem, which could then be given to an HTN planner to have it generate a plan to accomplish the goal task.  The LLM-generated methods and abstract tasks should embody good “control knowledge” for accomplishing the goal task in the domain, similar to what a human engineer could produce.  Let’s call this the HTN control knowledge generation problem. 
 + 
 +In this project, after learning the basics of HTN planning, the student will first adapt some existing HTN planning problems to be used as benchmarks in evaluating the ability of LLMs to solve the HTN control knowledge generation problem, essentially by deleting some or all of their methods and abstract tasks.  The student will also engineer prompts to try get the LLM to produce good quality solutions to the HTN control knowledge generation problem.  The performance of the LLM will then be evaluated in some experiments using the benchmark problems.   As part of this, the student will feed the LLM-generated HTN planning problems to an HTN planner and see if it can generate plans to solve them, providing one measure of correctness. 
 + 
 + 
 +**Required skills or prerequisites:**   
 +  - EECS 3401 
 +  - Python 
 + 
 +**Recommended skills or prerequisites:** 
 +  - basic knowledge of AI planning techniques and PDDL 
 +  - some experience using large language models 
 + 
 +**Instructions:** 
 +Contact instructor by email 
 + 
 +---- 
 + 
 + 
 + 
 + 
 +==== TEMPLATE ENTRY 10  - PUT PROJECT TITLE HERE====
  
 **[added YYYY-MM-DD]** **[added YYYY-MM-DD]**
Line 435: Line 587:
 **Instructions:** **Instructions:**
 //state how you wish to receive inquiries of interest// //state how you wish to receive inquiries of interest//
 +
 +
 ---- ----
  
-==== TEMPLATE ENTRY 10  - PUT PROJECT TITLE HERE====+ 
 +==== TEMPLATE ENTRY 11  - PUT PROJECT TITLE HERE====
  
 **[added YYYY-MM-DD]** **[added YYYY-MM-DD]**
Line 462: Line 617:
  
 ---- ----
- 
  
  
  
2024-25/fall/f24-listing.1724798559.txt.gz · Last modified: 2024/08/27 22:42 by hkh
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki