Recsperts - Recommender Systems Experts

#4: Adversarial Machine Learning for Recommenders with Felice Merra

Feb 23, 2022

Felice Merra, an applied scientist at Amazon, discusses Adversarial Machine Learning in Recommender Systems. Topics include perturbing data and model parameters, defense strategies, motivations for attacks, and privacy-preserving learning. The goal is to make systems more robust against potential attacks. They also touch on the challenges of robustifying multimedia recommender systems.

01:09:17

Creator website

Episode guests

Felice Merra

AI Summary

AI Chapters

Episode notes

Podcast summary created with Snipd AI

Quick takeaways

The impact of catalog size and user quantity on the vulnerability of recommender systems to attacks is crucial.

Safeguarding visual elements in recommender systems against adversarial attacks is essential for maintaining recommendation accuracy.

Defending against white box attacks targeting model parameters is vital for preserving the integrity of recommendation algorithms.

Deep dives

Effect of Large Catalogs on Model Attacks

Having a very huge catalog with few users makes it challenging to perform an attack on a recommender model. Conversely, a small catalog with many users makes it easier to influence the recommended items. This illustrates the impact of catalog size and user quantity on the vulnerability of a system to attacks.

Introduction

2min

Enhancing Security in Recommender Systems Through Adversarial Machine Learning

23min

Exploring Robustness Measurement in Recommender Systems

2min

Manipulating Image Embeddings for Recommender System Ranking

16min

Exploring User Meta Database Approaches and Their Influence on Recommendations

2min

Protecting Recommender Systems from Adversarial Attacks

21min

Closing Remarks on the Importance of Diverse Perspectives in Recommender Systems

4min

In episode four my guest is Felice Merra, who is an applied scientist at Amazon. Felice obtained his PhD from Politecnico di Bari where he was a researcher at the Information Systems Lab (SisInf Lab). There, he worked on Security and Adversarial Machine Learning in Recommender Systems.

We talk about different ways to perturb interaction or content data, but also model parameters, and elaborated various defense strategies.
In addition, we touch on the motivation of individuals or whole platforms to perform attacks and look at some examples that Felice has been working on throughout his research.
The overall goals of research in Adversarial Machine Learning for Recommender Systems is to identify vulnerabilities of models and systems in order to derive proper defense strategies that make systems more robust against potential attacks.
Finally, we also briefly discuss privacy-preserving learning and the challenges of further robustification of multimedia recommender systems.

Felice has published multiple papers at KDD, ECIR, SIGIR, and RecSys. He also won the Best Paper Award at KDD's workshop on Adversarial Learning Methods.

Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.

Links from this Episode:

Felice's Website
Felice Merra on LinkedIn and Twitter
Adversarial Machine Learning in Recommender Systems (PhD Thesis Final Presentation)
Workshop on Adversarial Personalized Ranking Optimization at ACM KDD 2021 (awarded Best Paper)
Adversarial Recommender Systems: Attack, Defense, and Advances (chapter in 3rd edition of Recommender Systems Handbook)
Information Systems Lab (SisInf Lab)

Thesis and Papers:

General Links:

Follow me on Twitter: https://twitter.com/LivesInAnalogia
Send me your comments, questions and suggestions to marcel@recsperts.com
Podcast Website: https://www.recsperts.com/

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.

Recsperts - Recommender Systems Experts

#4: Adversarial Machine Learning for Recommenders with Felice Merra

Episode guests

Podcast summary created with Snipd AI

Quick takeaways

Deep dives

Effect of Large Catalogs on Model Attacks

Adversarial Attacks on Visual Content in Recommender Systems

Consequences of Model Parameter Attacks

Need for Comprehensive Model Defense Strategies

Challenges in Reinforcement Learning and Multi-Modal Recommender Models

Remember Everything You Learn from Podcasts