Mahsa Khosh

I am an NLP/ML researcher at Son Corporate Group, develping machine learning and natural language processing to address practical challenges in the industry. I have completed my Master's degree in CS, specializing in Artificial Intelligence, from Amirkabir University of Technology. During my academic journey, Under the supervision of Professor Ebadzadeh, my research thesis explores the cognitive modeling of the human brain. Currently as a predoctoral fellow with the Fatima Fellowship, I collaborate with the UC Santa Barbara Natural Language Processing Group, under the guidance of my mentor, Michael Saxon. Our focus is on developing robust evaluation metrics for text-to-image models, crucial for achieving more faithful and consistent assessments. This effort plays a pivotal role in enhancing the performance and reliability of these models.

Over several years of industry experience, I've developed and applied numerous ML and NLP models to tackle real-world problems. This has resulted in gaining valuable insights and enhancing my skills. Now, after a decade of devoted work in the industry, my passion for returning to academia is stronger than ever, leading me to pursue a Ph.D. degree. The practical experiences gathered during these ten years have not only refined my skills but also sparked a genuine curiosity and a desire for further intellectual exploration. I'm motivated to pursue a Ph.D. because I'm eager to make meaningful contributions to the academic community, thoroughly explore unresolved and open problems, and share knowledge.

My main research goal is to enhance the reasoning abilities of large language models (LLMs) and evaluate their performance. One key challenge is improving how LLMs understand and process information in a manner that allows them to reason effectively. Current models often struggle with complex reasoning tasks, limiting their overall effectiveness. By addressing this challenge, I aim to develop innovative solutions that enable LLMs to excel in reasoning tasks across diverse domains. My research goals also extend to enhancing the multimodal capabilities of LLMs. In addition to improving reasoning abilities, I aim to develop solutions that enable LLMs to effectively process and understand data from various modalities including text, images, and other forms of information. By addressing this challenge, I seek to equip LLMs with versatility to integrate and reason across different types of data, thereby advancing their overall performance and applicability in real-world scenarios.

Research Interests:

• Large Language Models
• Vision-Language Models
• Commonsense Reasoning
• Multimodal AI