Hi everyone,
Long time no see!, we are excited to announce that we are back at organizing Data Science presentations for the benefit of the whole professional, academic and enthusiastic community!
We have two interesting speakers lined up for this coming June! We’ll meet up at 6:00 pm on Wednesday June 8th, and kick off talks shortly after.
Pizza and refreshments will be provided
***
Space is limited so be sure to RSVP. Also please cancel your RSVP in case you find you are not coming, regardless of how close to the meetup date you find out!
***
1.- Generating Refuted Claims and the Importance of Critical Statements in NLP Pipelines
Speaker: Domenic Rosati, Senior Research Scientist, scite.ai
Bio: Domenic’s research focuses on natural language understanding and reasoning in scientific texts and how to improve scientific discovery and evaluation with AI.
Abstract: With the advent of Large Language Models (LLMs) like GPT3 that perform at almost human levels without any fine tuning on many NLP (Natural Language Processing) benchmarks, a new line of research has emerged trying to assess the safety of these models by assessing the factuality and faithfulness of their outputs
Unfortunately, LLMs have been shown to be prone to “hallucination” that is they make up claims that are either irrelevant to the task at hand or unfactual. This prevents LLMs from being used in safety critical domains such as Scientific and Medical NLP.
However, most work has focused on a paradigm of verification of outputs (Gophercite, WebGPT, ect.) which has been shown to improve factuality benchmarks only slightly.
This talk will review work on improving factuality of LLMs by focusing on the refutation of outputs and why generating testable and critical statements from a model’s outputs is a useful paradigm for improving safe and interpretable deployment of LLMs.
2.- An Intro to Data for Good
Speaker: Michelle Sinville, Community Lead, Data for Good Maritimes
Bio: Michelle is an analytical thinker who enjoys using data to tell a story to a variety of audiences. She is an energetic and detail-oriented project scheduling consultant, specializing in biotech and pharmaceutical projects for Turner & Townsend. Michelle is also committed to building a thriving community through various volunteer initiatives.
Abstract: Are you looking for a way to use your data science skills for a good cause? Data for Good Maritimes is a branch of Data for Good, a Toronto-based non-profit that processes the data of other non-profits for free as a charitable service. Michelle will present a number of case studies from our recent projects, as well as an overview of how we engage with local nonprofits
See you there!
Register
