EMNLP2026Findings
01
About Me
I'm Baktash Ansari, a PhD student in Computer Science & Engineering at Michigan State University, in the HLR Lab advised by Prof. Parisa Kordjamshidi.
I work on natural language processing, vision-language modeling, and neuro-symbolic reasoning. I like figuring out how AI models understand language and images, and where their reasoning breaks down. I also enjoy building things end to end, from research code to full websites.
Research Interests
- Natural Language Processing
- Vision-Language Modeling
- Neuro-Symbolic Reasoning
- LLM Reasoning & Theory of Mind
- Spatial Reasoning
- Medical Imaging
02
Education & Journey
Aug 2026 — Present
PhD, Computer Science & Engineering
Michigan State University · East Lansing, MI
HLR Lab (Heterogeneous Learning and Reasoning), advised by Prof. Parisa Kordjamshidi.
Sep 2025 — Jun 2026
Research & Coursework, Computer Science GPA 3.97 / 4.0
University of Washington · Bothell, WA
Conducted research on retrieval & LLM reasoning as a Research Assistant and completed courses in AI and machine learning.
Oct 2020 — Jul 2024
BS, Computer Engineering GPA 3.92 · Rank #1 at graduation
Iran University of Science & Technology · Tehran
Top student of the department at the time of graduation. Specialization in AI + Machine Learning.
WebConf2026Workshop
ToxiTwitch: Toward Emote-Aware Hybrid Moderation for Live Streaming Platforms
CVPR2025Workshop
Illusory VQA: Benchmarking and Enhancing Multimodal Models on Visual Illusions
PDF ↗NAACL2024Workshop
BAMO at SemEval-2024 Task 9: BRAINTEASER
PDF ↗CLEF2024Workshop
Bilingual Sexism Classification: Fine-Tuned XLM-RoBERTa & GPT-3.5 Few-Shot Learning
04
Research Experience
PhD Researcher, HLR Lab
Aug 2025 — PresentMichigan State University · East Lansing, MI (remote Aug 2025 – Aug 2026, in person Aug 2026 – Present)
- Advisor: Dr. Parisa Kordjamshidi.
- Working on abductive reasoning using neuro-symbolic reasoning with LLMs.
- Paper under review.
Summer Research Intern
Summer 2026University of Washington · Bothell, WA
- Benchmarked 6 retinal foundation models on 17 diagnostic tasks for robustness to imaging artifacts and color removal.
- Showed that training on clean and artifact-corrupted images, with linear probing or end-to-end fine-tuning, improves accuracy on both corrupted and clean images, and that these gains transfer only partly to unseen artifacts.
- Paper under review.
Research Assistant
Mar 2025 — Jun 2026University of Washington · Bothell, WA
- Fine-tuned retrieval models (Contriever) with TopicTune, boosting Llama & GPT accuracy on open-domain Reddit/Lemmy data by 10%.
- Applied reinforcement learning with LLM feedback for retrieval optimization and generation.
- Implemented prompt-tuning on DeepSeek & Llama for toxicity detection in Twitch chat.
- Built multi-GPU PyTorch pipelines for inference, analysis & evaluation.
Research Assistant
Jul 2023 — Aug 2024Iran University of Science & Technology · Tehran
- Built 3 pipelines (Multi-Agent Debate, Chain-of-Thought, Fine-tuning) for SemEval-2024 Task 9, reaching 85% accuracy.
- Developed sexism-detection methods: 4th in Task 1, 2nd in Task 2 at CLEF 2024 EXIST.
- Generated 4 image datasets (4,000+ samples each) with Stable Diffusion to benchmark multimodal models (GPT-4o, Gemini, CLIP, BLIP).
05
Selected Projects
Hamming Embedding Image Search
Bag-of-Features with Hamming Embedding & Weak Geometric Consistency in C++. Achieved 63.46% mAP on Oxford, a 31% lift over baseline BoF.
Movie Revenue Prediction
End-to-end ML pipeline with 60+ engineered features across 17 models. Best: Extra Trees (R²=0.738), beating XGBoost & deep nets.
Persian Emotion Detection
Fine-tuned ParsBERT & XLM-RoBERTa Large on ArmanEmo; published models on Hugging Face. Multimodal classification with MVSA + ResNet.
BAMO @ SemEval 2024
Led the team on lateral-thinking reasoning: fine-tuned transformers, chain-of-thought prompting, and multi-agent debate (ReConcile).
Fuzzy Pendulum Control
Top-performing fuzzy control system for inverted-pendulum stabilization in a Computational Intelligence course.
TrekDestiny: Travel Platform
Led 3 devs building a travel web app with auth, real-time chat, notifications & blog. React + Tailwind, Scrum, CI/CD.
06
Technical Skills
Languages
ML / AI
Tools & Data
Honors & Awards
NSF NRT WaterCube Fellowship
National Science Foundation Research Traineeship, Michigan State University · 2026
Top Student in Computer Engineering
Ranked #1 of 102 students at the time of graduation, IUST · 2024
National Entrance Exam (Konkur)
Ranked 523rd, top 0.3% of 155,000+ · 2020
07
Let's Connect
Open to research collaborations, PhD discussions, and interesting problems in NLP, vision-language modeling, and retrieval.
[email protected]