Cost-sensitive Learning for Utility Optimization in Online - PowerPoint PPT Presentation

Sep 02, 2022 •483 likes •713 views

Cost-sensitive Learning for Utility Optimization in Online Advertising Auctions Flavian Vasile (Criteo) Damien Lefortier (Facebook) Olivier Chapelle (Google) Agenda Context Online & Offline Metrics Utility Optimization

Cost-sensitive Learning for Utility Optimization in Online Advertising Auctions Flavian Vasile (Criteo) Damien Lefortier (Facebook) Olivier Chapelle (Google)
Agenda • Context • Online & Offline Metrics • Utility Optimization • Online & Offline Results
Context (1) • Online Advertising Auctions for Display Advertising; 4 types of players: • The auction house: RTB platform, • Demand: the advertiser, • Supply: the publisher, • Bidder: Criteo.
Context (2) • Most people optimize for deep-funnel events and use a conversion rate (CR) prediction model. We focus on this aspect here.
Agenda • Context • Online & Offline Metrics • Utility Optimization • Online & Offline Results
Online Metrics • Conversions are different (e.g., sock vs. car) so we need to weight them by (some flavor of) CPA = Cost / #Conversions.
Offline Metrics
Agenda • Context • Online & Offline Metrics • Utility Optimization • Online & Offline Results
Utility Loss • Defined as the opposite of the Utility: • Non-convex; very hard.
Utility Loss and Log Loss (NLL) • We analyze the Utility loss when c is close to our bid pv. • We assume conversion probabilities are small (p << 1). => The derivatives are approximately equal, up to a factor v
Toy Example • Two advertisers with different CPAs (5 and 50) and CR (1% and 0.1%).
Method • We propose to optimize for WNLL to improve our bidder’s performance. • We use L-BFGS for learning.
Impact on Regularization • We propose the following heuristic to take weights into account:
Agenda • Context • Online & Offline Metrics • Utility Optimization • Online & Offline Results
Offline Setup • We use a public Criteo dataset for our experiments.
Offline Results – Weights
Offline Results – Lambda
Offline Results – High/low CPA
Online Results • The A/B test was done on more than 1 Billion ad displays, on world- wide traffic. Our change resulted in a +2% lift in ROI. • We observed significant savings in display cost + an increase in sales performance for the advertisers, especially on the campaigns with high CPA and low number of sales.
Conclusion • Weighted Log Loss allows to get closer to both offline and online metrics in the context of online advertising auctions.
Thanks!

Download Presentation

Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

C Cost-Sensitive Active t S iti A ti Visual Category Learning g y g Sudheendra

C Cost-Sensitive Active t S iti A ti Visual Category Learning g y g Sudheendra Vijayanarasimhan j y Kristen Grauman University of Texas at Austin y Learning visual categories Current category m odels d l Annotators wall koala

464 views • 29 slides

Cost-aware Pre-training for Multiclass Cost-sensitive Deep Learning Yu-An Chung 1 Hsuan-Tien Lin 1

Cost-aware Pre-training for Multiclass Cost-sensitive Deep Learning Yu-An Chung 1 Hsuan-Tien Lin 1 Shao-Wen Yang 2 1 Dept. of Computer Science and Information Engineering National Taiwan University, Taiwan 2 Intel Labs Intel Corporation, USA

526 views • 17 slides

Learning the Privacy-Utility Trade-off with Bayesian Optimization Borja Balle Joint work with B.

Learning the Privacy-Utility Trade-off with Bayesian Optimization Borja Balle Joint work with B. Avent, J. Gonzalez, T. Diethe and A. Paleyes Utility Privacy <latexit

532 views • 35 slides

Cost-Sensitive Batch Mode Active Learning: Designing Astronomical Observation by Optimizing

Cost-Sensitive Batch Mode Active Learning: Designing Astronomical Observation by Optimizing Telescope Time and Telescope Choice Xide Xia Advisor: Pavlos Protopapas Finale Doshi-Velez Planet Nebula Constellation Star cluster Galaxy

183 views • 17 slides

Utility Cost of Capital A Markets (and Judgment) Based Approach to Setting Utility Cost of

Utility Cost of Capital A Markets (and Judgment) Based Approach to Setting Utility Cost of Capital Shawn C. Allen, CFA, CMA, MBA, P. Eng. This material represents the personal views of 1 the author Contents Definition of Capital and

1.32k views • 113 slides

Cost-sensitive Dynamic Feature Selection e III 1 and Jason Eisner 2 He He 1 , Hal Daum 1

Cost-sensitive Dynamic Feature Selection e III 1 and Jason Eisner 2 He He 1 , Hal Daum 1 University of Maryland, College Park 2 Johns Hopkins University June 30, 2012 He He, Hal Daum e III and Jason Eisner Cost-sensitive Dynamic Feature

641 views • 46 slides

Optimization Optimization Goal: Find the minimizer ! that minimizes the objective (cost)

Optimization Optimization Goal: Find the minimizer ! that minimizes the objective (cost) function # ! : & Unconstrained Optimization # ! = min ! # ! Constrained Optimization # ! = min ! # ! s.t. , ! = - Equality

882 views • 38 slides

Selective sampling algorithms for cost-sensitive multiclass prediction Alekh Agarwal Microsoft

Selective sampling algorithms for cost-sensitive multiclass prediction Alekh Agarwal Microsoft Research Alekh Agarwal Selective sampling for multiclass prediction Why active learning? Standard setting - receive randomly sampled examples

769 views • 53 slides

Cost Utility Analysis Subgroup Brian Custer ISBT WP-TTID Cancun, Mexico July 8, 2012 Global

Cost Utility Analysis Subgroup Brian Custer ISBT WP-TTID Cancun, Mexico July 8, 2012 Global risk assessment and cost utility of blood safety interventions development of a web- based application and multi-country analysis framework

700 views • 25 slides

3. Preference Learning Techniques a. Learning Utility Functions b. Learning Preference

AGENDA 1. Preference Learning Tasks 2. Performance Assessment and Loss Functions 3. Preference Learning Techniques a. Learning Utility Functions b. Learning Preference Relations c. Structured Output Prediction d. Model-Based Preference

583 views • 39 slides

Learning Outcomes I understand the difference between level- sensitive and edge-sensitive

13.1 13.2 Learning Outcomes I understand the difference between level- sensitive and edge-sensitive Unit 13 I understand how to create an edge-triggered FF from 2 latches Sequential Logic Constructs 13.3 13.4 Sequential Logic

457 views • 12 slides

Optimization (Introduction) : IR IR f ( x ) ID Optimization " 112 FCI ) : IR NE

Optimization (Introduction) : IR IR f ( x ) ID Optimization " 112 FCI ) : IR NE Goal: Find the minimizer ! that minimizes the objective (cost) function " ! : " - fatal * Unconstrained Optimization x :

408 views • 23 slides

A Cost-Sensitive Adaptation Engine for Server Consolidation of Multitier Applications Gueyoung

A Cost-Sensitive Adaptation Engine for Server Consolidation of Multitier Applications Gueyoung Jung, Calton Pu Georgia Institute of Technology Kaustubh Joshi, Matti Hiltunen, Richard Schlichting AT&T Labs Research Context Cloud

475 views • 26 slides

A6: Sensitive Data Exposure A6 Sensitive Data Exposure Sensitive data stored or transmitted

A6: Sensitive Data Exposure A6 Sensitive Data Exposure Sensitive data stored or transmitted insecurely Failure to protect all sensitive data Usernames, passwords, password hashes, credit-card information, identity info Session

415 views • 17 slides

Evaluation Of Post-Hoc Optimization Constraints Under Altered Cost Functions Presentation of

Introduction Operator-Counting Cost-Altered Post-Hoc Optimization Constraints Experiments Conclusion References Evaluation Of Post-Hoc Optimization Constraints Under Altered Cost Functions Presentation of Masters Thesis Andreas Th

628 views • 30 slides

COOL: Common Optimization and Operation framework based on network utiLity theory for 5G

1 IEEE 5G Greece Summit COOL: Common Optimization and Operation framework based on network utiLity theory for 5G technologies & IoT Symeon Papavassiliou National Technical University of Athens (NTUA) School of Electrical and Computer

143 views • 12 slides

CS490W: Web I nformation Systems Some core concepts of I R Information CS-490W Need Web

CS490W: Web I nformation Systems Some core concepts of I R Information CS-490W Need Web Information Systems Representation Course Review Query Retrieval Model Indexed Objects Luo Si Retrieved Objects Department of Computer Science

177 views • 7 slides

HOME Essentials January 12, 2020 Presented by: Monte Franke MLFranke@aol.com Slide 2 HOMEin 4

HOME Essentials January 12, 2020 Presented by: Monte Franke MLFranke@aol.com Slide 2 HOMEin 4 hours? History & overview Eligible activities & costs Program partners & roles Administration Program

1.26k views • 89 slides

Arbitration vs. Litigation November 15, 2017 Choosing Your Dispute Resolution Method Wisely

Arbitration vs. Litigation November 15, 2017 Choosing Your Dispute Resolution Method Wisely James Tancula Miles Robinson Partner Partner +1 312 701 7900 +44 20 3130 3974 jtancula@mayerbrown.com miles.robinson@mayerbrown.com Speakers

198 views • 18 slides

Bartering for Free Information: Implications for GDP and Productivity Leonard Nakamura, Jon

Bartering for Free Information: Implications for GDP and Productivity Leonard Nakamura, Jon Samuels and Rachel Soloveichik Introduction Free content isnt currently included with final expenditures in measured GDP.

319 views • 20 slides

Bayesian Networks C H A P T E R 1 4 H A S S A N K H O S R A V I S P R I N G 2 0 1 1

Bayesian Networks C H A P T E R 1 4 H A S S A N K H O S R A V I S P R I N G 2 0 1 1 Definition of Bayesian networks Representing a joint distribution by a graph Can yield an efficient factored representation for a joint

965 views • 67 slides

Time Series A time series is a collection of data obtained by observing a response variable at

ST 430/514 Introduction to Regression Analysis/Statistics for Management and the Social Sciences II Time Series A time series is a collection of data obtained by observing a response variable at various times. Examples The monthly Case-Shiller

351 views • 8 slides

$O \ S' I q \ \ ;.. { \ I ) (r. - \ \ Y} ls I -{ o\) C; \ \0 >- --r)$

O \ S' I q \ \ ;.. { \ I ) (r. - \ \ Y} ls I -{ o\) C; \ \0 >- --r)

s 5-5 \0 p -s LJ \) J 1 -J 'a s \) o .J o .6 \ \) -)- ,: v - \\ ! \ FI .6 \) \n s \s v \ -? \) \-) s -!_ o - s ( 5 \, -:o S I --1 O \ S' I q \ \ ;.. { \ I ) (r. - \ \ Y} ls I -{ o\)

512 views • 25 slides

Introduction to video reverse engineering Vittorio Giovara Brussels 2016-01-29 FOSDEM - Open

Introduction to video reverse engineering Vittorio Giovara Brussels 2016-01-29 FOSDEM - Open Media CC-BY-SA 1 About me Libav/FFmpeg developer (~10 decoders) VideoLAN association member First known use of x264 in broadcasting

579 views • 29 slides