Final artifacts
Research reports
These reports can be cited as “Dickinson College DNU Lab independent research report, spring 2026”.
- Azad, Pranav. Evaluating Data Representation and Chain-of-Thought Reasoning for Arithmetic Tasks in Character-Level Language Models
- Goda, Youssif; Lee, John; Sanchez, Maximo. Hardware Comparison on Tiny Shakespeare (Character-Level nanoGPT)
- Ifrad, Mubashshir Alam. Transformer Reasoning Research
- Jadoon, Tayyaba. Reasoning Without Generalization: A Comparison of Direct, Chain-of-Thought, and Scratchpad Training Paradigms on Out-of-Distribution Arithmetic
- Kapa, Hemanth Surya Ganesh. Latent Chain-of-Thought via Looped Transformers
- Muminov, Aziz. Cross-Step Activation Mixing Has a Depth-Dependent Design Tradeoff: Preliminary Results on Multi-Digit Addition
- Nguyen, Phu. Teaching Arithmetic to a Small Language Model: A Chain-of-Thought Training Study with NanoGPT
- Nguyen, Son. Transformer Algebra for Reverse-Then-Add Composition
- Patwary, Radiath Kamal. From Readings About LLM Reasoning and Some Research Experiments
- Pham, Lam. Optimizing nanoGPT for Multi-Digit Addition through Hardware, Algorithm, and MLOps Restructuring
- Piya, Rohan. Can Chain-of-Thought Improve Algorithmic Reasoning in Small Language Models?
- Saha, Niloy. What Tiny GPTs Can Represent and What They Actually Learn
- Shin, Aaron. Target Masking in a Tiny Transformer: Higher Ceiling, Not Faster Convergence
Note: Reports are provided as submitted by students and have not been validated by faculty. In this independent research course, students were encouraged to set their own goals and priorities for final deliverables. In some cases, students chose to share results informally with the instructors and other students, opting for a brief or unpolished final report. In other cases, students prioritized the production of a formal scientific report.
Posters
- Goda, Youssif; Lee, John; Sanchez, Maximo. GPU vs CPU Performance Analysis on Character-Level AI Models
- McClellan, Coby. DataFrame QA: Privacy-Preserving Data Analysis via Schema Metadata