dimacs logo color

Events

Event Calendar

« Previous Month

August 2026

Next Month »
Sun Mon Tue Wed Thu Fri Sat
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

Seminar Details

Auditing Black-box Models

  • Start Date: June 13, 2017
  • Event Start Time: 2:00 PM
  • Event End Time: 3:00 PM
  • Organizers: Lazaros Gallos
  • Seminar Series: REU Seminar
  • Presenter(s): Sorelle Friedler - Haverford College
  • Event Location: DIMACS Seminar room
  • Abstract: Machine learning models are becoming increasingly opaque to human examination, even to their designers. Understanding model predictions is broadly important across public policy, scientific hypothesis generation, and model optimization, and so there have been increasing calls for focuses on the accountability and transparency of machine learning. But how can we practically achieve accountability and transparency in the face of increasingly complex models? In this talk, we’ll discuss strategies for auditing black-box models when given access to their inputs and outputs. Focusing on the goal of quantifying the influence of each feature on the model’s predictions, we’ll consider techniques for identifying both direct and indirect influence. We’ll see these techniques applied to models from both experimental chemistry and social science applications, beginning to shed light on complex or even proprietary models.