Differences

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

--- 2026-27:fall [2026/06/18 23:34] – vashistp
+++ 2026-27:fall [2026/08/12 17:11] (current) – vashistp
@@ Line 33: / Line 33: @@
 ----
+==== AI for medical imaging ====
+**[added 2026-06-19]**
+**Course:**  {EECS4080/EECS4088/4070 (only if the student also takes 4080 or 4088 with Prof. Laleh)}
+**Supervisor:**  Laleh Seyyed-Kalantari
+**Supervisor's email address:**  lsk@yorku.ca
+**Project Description:**
+We are developing deep learning models for medical image diagnostics and evaluating their biases.
+**Required skills:**
+  - Machine learning courses
+  - Programing in Python and PyTorch
+  - Experience in deep learning and LLMs
+  - GPA in A, A+ range or demonstration of existing successful projects in machine learning. GPA below B+ will not be considered.
+I encourage students who are motivated for real, impactful research to apply. Those who may be willing to take the project and continue it to get to the publication stage. Upon successful research experience, students will be considered for NSERC USRA/LURA and Vector Institute Master Award nomination and Connected Minds graduate scholarship (in case they apply for the Master's program at York). So far, our team have had 100% award acceptance rate in diverse program for graduate students.
+**Recommended skills:**
+  - Experience in computer vision
+**Instructions:**
+If you would like to apply, please complete the [[https://docs.google.com/forms/d/e/1FAIpQLSdQk7ChC59SqY-dVYmNPfzkAJWWLpeugXmYrU-NK9-EN0GH8g/viewform | Responsible AI Lab's undergraduate research project form]].
+[[https://responsibleai.eecs.yorku.ca/index.html | Learn more about the Responsible AI lab]].
+==== Bias in Vision Language Models ====
+**[added 2026-06-19]**
+**Course:**  {EECS4080/EECS4088/4070 (only if the student also takes 4080 or 4088 with Prof. Laleh)}
+**Supervisor:**  Laleh Seyyed-Kalantari
+**Supervisor's email address:**  lsk@yorku.ca
+**Project Description:**
+ If you give a Vision Language Model to interpret a neutral image, the interpretation varies by the demographic that we mentioned in the image description. For example, if you ask it to interpret an image related to Western country, it focuses on the positive side, while for the same image, it describes the negative side if you say the image belongs to an African country. Similarly, we want to develop benchmarks to evaluate the bias of VLMs in multi-lingual settings, i.e., if the language is Persian, vs English or Chinese, etc., how will the biases change? We can adjust the project based on the native language of students who take the project.
+Required Skills: Looking for highly motivated students with a GPA in A, A+ range. Please send your CV, all your transcripts to date, and any prior experience (if any) with Python, PyTorch, deep learning, and LLMs. Please list all languages that you are familiar. For this particular project, please list if you have other native languages besides English.
+**Required skills:**
+  - Machine learning courses
+  - Programing in Python and PyTorch
+  - Experience in deep learning and LLMs
+  - GPA in A, A+ range or demonstration of existing successful projects in machine learning. GPA below B+ will not be considered.
+I encourage students who are motivated for real, impactful research to apply. Those who may be willing to take the project and continue it to get to the publication stage. Upon successful research experience, students will be considered for NSERC USRA/LURA and Vector Institute Master Award nomination and Connected Minds graduate scholarship (in case they apply for the Master's program at York). So far, our team have had 100% award acceptance rate in diverse program for graduate students.
+**Recommended skills:**
+  - Experience in computer vision
+**Instructions:**
+If you would like to apply, please complete the [[https://docs.google.com/forms/d/e/1FAIpQLSdQk7ChC59SqY-dVYmNPfzkAJWWLpeugXmYrU-NK9-EN0GH8g/viewform | Responsible AI Lab's undergraduate research project form]].
+[[https://responsibleai.eecs.yorku.ca/index.html | Learn more about the Responsible AI lab]].
+==== Dialect bias in LLMs ====
+**[added 2026-06-19]**
+**Course:**  {EECS4080/EECS4088/4070 (only if the student also takes 4080 or 4088 with Prof. Laleh)}
+**Supervisor:**  Laleh Seyyed-Kalantari
+**Supervisor's email address:**  lsk@yorku.ca
+**Project Description:**
+When an African American English (AAE) speaker writes ``I be working'' to express a habitual action, large language models (LLMs) often misinterpret, penalize, or ``correct'' this grammatically valid construction. This is ‘dialect preference bias’: the systematic tendency of LLMs to favor Standard American English (SAE) over equally valid language varieties like AAE, spoken by over 30 million people worldwide. The consequences are tangible. LLMs disproportionately misclassify AAE as toxic or harmful, respond to AAE speakers with more stereotyping and condescension, and exhibit degraded performance on downstream tasks when inputs are in non-standard dialects. As LLMs become embedded in high-stakes domains such as hiring, healthcare, and content moderation, dialect bias threatens to systematically disadvantage the millions who speak marginalized varieties of English.
+We want to further explore these biases for the African American dialect and the African dialect and develop for assessment benchmarks. Please list all languages that you are familiar. For this particular project, please list if you belong to the African American community or have African roots, or if you have another dialect (e.g. Indian English) that you would like to perform the study on that dialect.
+**Required skills:**
+  - Machine learning courses
+  - Programing in Python and PyTorch
+  - Experience in deep learning and LLMs
+  - GPA in A, A+ range or demonstration of existing successful projects in machine learning. GPA below B+ will not be considered.
+I encourage students who are motivated for real, impactful research to apply. Those who may be willing to take the project and continue it to get to the publication stage. Upon successful research experience, students will be considered for NSERC USRA/LURA and Vector Institute Master Award nomination and Connected Minds graduate scholarship (in case they apply for the Master's program at York). So far, our team have had 100% award acceptance rate in diverse program for graduate students.
+**Recommended skills:**
+  - Experience in computer vision
+**Instructions:**
+If you would like to apply, please complete the [[https://docs.google.com/forms/d/e/1FAIpQLSdQk7ChC59SqY-dVYmNPfzkAJWWLpeugXmYrU-NK9-EN0GH8g/viewform | Responsible AI Lab's undergraduate research project form]].
+[[https://responsibleai.eecs.yorku.ca/index.html | Learn more about the Responsible AI lab]].
 ==== Gamification of how we learn Discrete Mathematics ====
-**[added 2025-25-08]**
+**[added 2026-08-08]**
 **Course:** {EECS4480/4080/4088}
-**Project Description:** This project explores how game-based learning can make Discrete Mathematics more engaging and enjoyable for students. Topics in this course—such as logic, proofs, sets, functions, and number systems—are often challenging because they feel abstract and disconnected from everyday experience. The goal of this project is to design a learning experience that presents these ideas as a series of interactive “quests,” similar to the levels in an adventure game. Students will move through a visual map where each quest represents a key concept in Discrete Mathematics. For example, one quest may involve solving a puzzle to cross a bridge, while another may involve unlocking a gate by understanding logical statements. The project will focus on how storytelling, rewards, and interactive challenges can help students stay motivated and understand concepts more clearly. A simple digital prototype will be created—using design tools such as Figma or basic web interfaces—to demonstrate how the game experience would look and feel. The goal is to create a clear and engaging representation of how gamification could improve learning in Discrete Mathematics.
+**Project Description:** This project explores how game-based learning can make Discrete Mathematics more engaging and enjoyable for students. Topics in this course—such as logic, proofs, sets, functions, and number systems—are often challenging because they feel abstract and disconnected from everyday experience. The goal of this project is to design a learning experience that presents these ideas as a series of interactive “quests,” similar to the levels in an adventure game. Students will move through a visual map where each quest represents a key concept in Discrete Mathematics. For example, one quest may involve solving a puzzle to cross a bridge, while another may involve unlocking a gate by understanding logical statements. The project will focus on how storytelling, rewards, and interactive challenges can help students stay motivated and understand concepts more clearly. A digital game  has to be developed that is to be created—using design tools such as Figma and web interfaces—to demonstrate how the game experience would look and feel; and how the game can be used. The goal is to create a clear and engaging representation of how gamification could improve learning in Discrete Mathematics.
 **Required/Recommended skills or prerequisites:** Basic understanding of Discrete Mathematics concepts. Familiarity with simple design tools (e.g., PowerPoint, Figma, or Canva). Interest in UI/UX design or game design. Programming knowledge (e.g., JavaScript or Python). Creativity and interest in storytelling or visual design
@@ Line 133: / Line 225: @@
 ** Course:**  {EECS4080}
-** Supervisors:**  Pooja Vashith
+** Supervisors:**  Pooja Vashisth
 ** Supervisor's email address: ** vashistp@yorku.ca
@@ Line 139: / Line 231: @@
 ** Project Description: ** This project aims to uncover meaningful insights from EECS course evaluations by applying natural language processing (NLP) techniques to student feedback. While most universities collect large volumes of student comments in course evaluations, these are typically underused, especially when embedded in PDF files. Qualitative feedback is often reviewed manually or averaged superficially, leaving behind rich emotional and experiential data that could inform course improvement.
-The primary goal is to build a processing pipeline that extracts, cleans, and analyzes this feedback using both basic sentiment analysis tools (e.g., VADER) and advanced emotion classification models (e.g., GoEmotions). The emotional tone expressed in the feedback will be mapped to different course components such as the instructor, teaching assistant, assessments, and course content. NB: These are already separated in the evaluation structure.
+The primary goal is to advance an existing application. The current application is a processing pipeline that extracts, cleans, and analyzes this feedback using both basic sentiment analysis tools (e.g., VADER) and advanced emotion classification models (e.g., GoEmotions).
-By comparing the expressiveness and usefulness of simple versus fine-grained emotional analysis, this research will help determine which approaches are more effective at surfacing actionable insights. These insights will be visualized to highlight recurring patterns of sentiment or emotion across course components, such as whether students consistently express frustration about assessments or admiration for certain instructors.
+By comparing the expressiveness and usefulness of simple versus fine-grained emotional analysis, this research will help determine which approaches are more effective at surfacing actionable insights. These insights will be visualized to highlight recurring patterns of sentiment or emotion across course components, such as whether students consistently express frustration about assessments or admiration for certain instructors. The aim is to use and test the results and develop a strong research writing of publishable nature.
-This project is educational in nature as it equips the student with skills in text analytics, NLP tools, and data visualization while contributing to a broader understanding of how data-driven analysis can support evidence-based teaching and curriculum refinement in academic institutions.
+This project is an academic research in nature as it equips the student with skills in research paper writing, survey, text analytics, NLP tools, and data visualization while contributing to a broader understanding of how data-driven analysis can support evidence-based teaching and curriculum refinement in academic institutions.
 ** Required skills or prerequisites: **EECS 4412 or EECS4404
@@ Line 149: / Line 241: @@
 Data Analysis, Report Writing, Python programming, web app development, appetite for research
-** Instructions:** sen a CV, transcript, statement of interest, and skills to the instructor (Pooja).
+** Instructions:** send a CV, transcript, statement of interest, and skills to the instructor (Pooja).
+----
+==== Adaptive Routing for Cost-Efficient Autonomous Email Systems ====
+** [added 2026-08-08] **
+** Course:**  {EECS4088}
+** Supervisors:**  Pooja Vashisth
+** Supervisor's email address: ** vashistp@yorku.ca
+** Project Description: ** This project aims to implement Agentic system for an email assistant.
+This project designs and implements a multi-agent email processing system that routes incoming emails to different handlers based on email type, rather than always invoking an expensive large language model (LLM). The system classifies emails into categories such as acknowledgment, meeting request, FAQ, and emotional, and dispatches each to the cheapest appropriate handler — including a template engine, cache, retrieval module, calendar integration, and LLM fallback. A DistilBERT classifier is implemented as an upgrade to a rule-based baseline, enabling a direct comparison. The system is evaluated on routing accuracy, token cost, and response quality against multiple baselines, with the goal of demonstrating meaningful cost reduction without quality degradation.The aim is to use and test the results and develop a strong research writing of publishable nature.
+This project is an academic research in nature as it equips the student with skills in research paper writing, survey, text analytics, NLP tools, and data visualization.
+** Required skills or prerequisites: **
+Data Analysis, Report Writing, Python programming, web app development, appetite for research
+** Instructions:** send a CV, transcript, statement of interest, and skills to the instructor (Pooja).
 ----