Exploring Cutting Edge AI Research at ICML 2024 with Capital One
The 41st International Conference on Machine Learning (ICML) took place in Vienna, Austria this year, attracting AI enthusiasts from all over the world, including members of the Capital One AI research team. ICML serves as a platform for attendees to delve into the latest advancements in AI, machine learning, and related fields such as generative AI, natural language processing, computer vision, and reinforcement learning.
For Capital One, ICML presented an opportunity to unite our team, attend expert sessions, and collaborate with industry peers to push the boundaries of AI innovation. Amidst all the knowledge sharing, our team also took the time to showcase the impactful work we are doing in the AI space while enjoying the sights of the city.
As the conference unfolded, attendees engaged in a myriad of activities including tutorials, keynotes, lightning talks, and poster sessions, creating an atmosphere of energizing exchange and ideation for the future.
Several presentations at ICML captured our attention, shedding light on critical advancements in the AI domain. Our team was particularly intrigued by the discussions around training data for large language models (LLMs) and the methods for supervising and evaluating LLM outputs.
Training Data for Large Language Models (LLMs)
Papers such as “Understanding Finetuning for Factual Knowledge Extraction” and “Memorization through the lens of curvature of loss function around samples” provided valuable insights into how models learn from data and the implications of overfitting. Additionally, a tutorial on the “Physics of Language Models” offered a deep dive into the workings of LLMs and effective data preparation techniques.
Supervising and Evaluating LLM Outputs
Discussions on “Weak-to-Strong Generalization” and “Debating with More Persuasive LLMs Leads to More Truthful Answers” showcased innovative approaches to evaluating complex AI systems. A tutorial on the “Lessons from the Trenches on Reproducible Evaluation of Language Models” outlined best practices for evaluating LLM outputs.
In addition to knowledge sharing, Capital One hosted an engaging dinner conversation with academic and industry leaders to explore advancements in Generative AI and the importance of building ethical and reliable systems. The event fostered insightful discussions on the evolution of AI and the challenges posed by human-machine content collaboration.
Capital One associates actively participated in the conference, engaging with students and professionals to discuss industry challenges and share insights. The experience at ICML 2024 has equipped our team with fresh perspectives and newfound inspiration to continue driving AI innovation in the finance sector.
As we reflect on our time at ICML, we are excited to leverage the knowledge gained to further enhance our AI capabilities at Capital One and contribute to shaping the future of financial services through transformative AI solutions.
If you’re passionate about AI and eager to make a difference in the finance industry, explore Applied Research jobs at Capital One and be part of a dynamic team shaping the future of banking.