The Geisel School of Medicine Events Calendar
The Geisel School of Medicine Events - ( Subscribe )
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Saeed Hassanpour (Geisel) will present “Large Language Models in Healthcare: Promise and Pitfalls,” and Sharmistha Sidkar (Tuck) will present “Neither a Picasso Nor a Da Vinci: A Multi-modal Model for Pricing of Novice Artwork.” The seminar is noon-1pm in Volanakis classroom at Tuck or by Zoom: https://dartmouth.zoom.us/j/97590036003?pwd=UWw0Z2U1ZzZjRExReFJ3dlIyMkNIdz09
Saeed Hassanpour (Geisel)
Title: Large Language Models in Healthcare: Promise and Pitfalls
Abstract: This talk delves into the dynamic world of Large Language Models (LLMs) in healthcare, highlighting their revolutionary applications and inherent challenges. We will explore LLMs' capabilities in summarizing medical literature, enhancing clinical documentation, aiding clinical decision-making, and improving patient communication and empathy, particularly in low-resource settings. The talk will also address critical concerns such as the risk of generating incorrect information and the limitations in current medical knowledge coverage. By examining the balance between these groundbreaking technologies and their potential pitfalls, this presentation aims to shed light on how LLMs can be optimally utilized to advance healthcare while maintaining safety and accuracy in clinical practice.
Sharmistha Sikdar (Tuck)
Title: Neither a Picasso Nor a Da Vinci: A Multi-modal Model for Pricing of Novice Artwork
Abstract: Art pricing is a difficult problem, especially in the novice art market, where the artists' reputation is unknown, and artwork is purchased for consumption. Online advisory sites use rudimentary measures like artwork dimensions, material costs, and artist's labor hours to suggest prices, although extant literature reports buyer's willingness to pay (WTP) to be the most important factor in art price determination. Novice art sold on online platforms like Etsy is listed with artwork and artist features, textual descriptions, and visual images that might affect a buyer's WTP. Using scraped data of sold artwork from Etsy, we build a scalable, multi-modal deep learning model to predict customer WTP. Our model reveals the relative importance of characteristics related to the artwork's durability, exclusivity, and usage ease over the textual descriptions and visuals, in predicting WTP. Thus, novice art pricing has similarities but also notable differences from established artwork pricing. Our WTP experiment shows that our model predicts respondent WTP better than the online advisory sites. We develop an online application for emerging artists and platforms like Etsy to generate price recommendations that are closer to customer WTP for novice artwork.
Saeed Hassanpour (Geisel)
Title: Large Language Models in Healthcare: Promise and Pitfalls
Abstract: This talk delves into the dynamic world of Large Language Models (LLMs) in healthcare, highlighting their revolutionary applications and inherent challenges. We will explore LLMs' capabilities in summarizing medical literature, enhancing clinical documentation, aiding clinical decision-making, and improving patient communication and empathy, particularly in low-resource settings. The talk will also address critical concerns such as the risk of generating incorrect information and the limitations in current medical knowledge coverage. By examining the balance between these groundbreaking technologies and their potential pitfalls, this presentation aims to shed light on how LLMs can be optimally utilized to advance healthcare while maintaining safety and accuracy in clinical practice.
Sharmistha Sikdar (Tuck)
Title: Neither a Picasso Nor a Da Vinci: A Multi-modal Model for Pricing of Novice Artwork
Abstract: Art pricing is a difficult problem, especially in the novice art market, where the artists' reputation is unknown, and artwork is purchased for consumption. Online advisory sites use rudimentary measures like artwork dimensions, material costs, and artist's labor hours to suggest prices, although extant literature reports buyer's willingness to pay (WTP) to be the most important factor in art price determination. Novice art sold on online platforms like Etsy is listed with artwork and artist features, textual descriptions, and visual images that might affect a buyer's WTP. Using scraped data of sold artwork from Etsy, we build a scalable, multi-modal deep learning model to predict customer WTP. Our model reveals the relative importance of characteristics related to the artwork's durability, exclusivity, and usage ease over the textual descriptions and visuals, in predicting WTP. Thus, novice art pricing has similarities but also notable differences from established artwork pricing. Our WTP experiment shows that our model predicts respondent WTP better than the online advisory sites. We develop an online application for emerging artists and platforms like Etsy to generate price recommendations that are closer to customer WTP for novice artwork.