Alexander Goslin Email and Phone Number

• Enhancing a data analysis library through code refactoring, improving maintainability and readability.• Integrated configuration files to minimize hard-coding, allowing for easy adjustments to parameters, and enhancingsoftware flexibility and maintainability.• Identifying and resolving software bugs, creating more stable and error-free software.

• Conducted research at the University of Maryland at the intersection of generative models, computervision, natural language processing, fairness, and reinforcement learning.• Created dataset using Stable Diffusion for training a highly accurate gender classifier employing state ofthe art computer vision models. Used gender classifier to develop novel reward function for mitigatinggender-bias in stable diffusion.• Applied reward function to stable diffusion using recently-published technique for training diffusionmodels using reinforcement learning (DDPO), as well as HuggingFace Diffusers/Transformers.• Extended DDPO algorithm to increase rewards by exclusively modifying prompt representations,increasing convergence speed and image quality.

• Led engaging discussions for 30 students, enhancing their grasp of discrete math concepts.• Collaborated on coursework design, aligning assignments and exams with course objectives.• Graded coursework, offered tailored student support, and contributed to course discussions.