Agenda:
17:30-18:00: Networking
18:00-18:45: Bid optimization, by Alex Shtoff.
18:45-19:30: From Pixels to Prose: The Rise of Diffusion Language Models, by Shahar Gigi
Bid optimization
Only advertising plays a central role in funding most of our apparently free services and drives a large portion of the web economy. The ad we eventually see is typically selected using an auction mechanism, where a large number of ads participate with an associated bid, and the highest bidder wins. The bid optimization task deals with choosing the bid every time an auction is conducted, repeatedly, to optimize some desired performance goal, such as exposure, or return on investment, while keeping some portion of the advertiser's spend as the revenue of the advertising platform. Machine learning is the driving force behind most bid optimization algorithms, but models are deployed for this task in a unique manner and a unique set of constraints, that typically differ from most other machine learning applications we observe. In this talk we will discuss the uniqueness of bid optimization, the challenges it poses, and present several machine-learning driven approaches.
Alex Shtoff is a Principal Scientist at the AI-IR research center of the UAE Technology Innovation Institute (TII). Previously, he worked at Yahoo, mainly in the native advertising team. Alex holds a Ph.D in Operations Research from the Technion, and an M.Sc in Computer Science from Tel Aviv university.
From Pixels to Prose: The Rise of Diffusion Language Models
From Pixels to Prose: The Rise of Diffusion Language Models
While Autoregressive (AR) models like GPT dominate today's landscape, Diffusion Language Models (DLMs) are emerging as a formidable alternative. This lecture surveys the rapid evolution of DLMs, tracing their trajectory from the revolution in image generation to the latest breakthroughs in discrete text processing.
We will dive into the architectural shift from "next-token" prediction to iterative denoising, exploring how these models utilize bidirectional context to outshine AR models in complex reasoning and data-scarce environments.
Shahar Gigi s a senior data scientist at ZIM where he works mainly on classical ML for pricing and consumption modeling. Prior to that he worked in SamsungNext on computer vision models for smart devices and iot.
