Benjamin Heymann
@benhey.bsky.social
Researcher at Criteo AI lab
Game theory, optimisation and AI
Game theory, optimisation and AI
extended abstracts are on my website: benhey.github.io
Benjamin Heymann
A simple, whitespace theme for academics. Based on [*folio](https://github.com/bogoli/-folio) design.
benhey.github.io
September 19, 2025 at 2:48 PM
extended abstracts are on my website: benhey.github.io
The programme is here sites.google.com/view/consequ...
CONSEQUENCES 2025 - Programme
September 22nd 2025 - Room D3
11:30 – 11:35 Welcome by the organisers
11:35 – 12:30 Tutorial: Practical Bandits: An Industry Perspective
Bram van den Akker, Olivier Jeunen, Ben London and...
sites.google.com
September 19, 2025 at 2:48 PM
The programme is here sites.google.com/view/consequ...
📍Poster: 10am
🎤 Talk: 11:30am
📄 Paper: arxiv.org/abs/2409.03875
If you're at #AAMAS2025, come by and say hi to @sharky6000.bsky.social!
And let us know what you think! 🤝
6/6
🎤 Talk: 11:30am
📄 Paper: arxiv.org/abs/2409.03875
If you're at #AAMAS2025, come by and say hi to @sharky6000.bsky.social!
And let us know what you think! 🤝
6/6
Learning in Games with Progressive Hiding
When learning to play an imperfect information game, it is often easier to first start with the basic mechanics of the game rules. For example, one can play several example rounds with private cards r...
arxiv.org
May 23, 2025 at 1:33 PM
📍Poster: 10am
🎤 Talk: 11:30am
📄 Paper: arxiv.org/abs/2409.03875
If you're at #AAMAS2025, come by and say hi to @sharky6000.bsky.social!
And let us know what you think! 🤝
6/6
🎤 Talk: 11:30am
📄 Paper: arxiv.org/abs/2409.03875
If you're at #AAMAS2025, come by and say hi to @sharky6000.bsky.social!
And let us know what you think! 🤝
6/6
To our knowledge, this is the first use of information relaxation techniques from operations research in computational game theory.
We're excited to see how this perspective bridges communities!
#AAMAS2025 #ReinforcementLearning #MultiAgentSystems
5/6
We're excited to see how this perspective bridges communities!
#AAMAS2025 #ReinforcementLearning #MultiAgentSystems
5/6
May 23, 2025 at 1:32 PM
To our knowledge, this is the first use of information relaxation techniques from operations research in computational game theory.
We're excited to see how this perspective bridges communities!
#AAMAS2025 #ReinforcementLearning #MultiAgentSystems
5/6
We're excited to see how this perspective bridges communities!
#AAMAS2025 #ReinforcementLearning #MultiAgentSystems
5/6
🧠 The idea:
1️⃣ Train as if all information is visible
2️⃣ Penalize the agent for using “forbidden” info (hide it)
This enables fast learning of policies that respect information constraints.
Hence: Progressive Hiding. 4/6
1️⃣ Train as if all information is visible
2️⃣ Penalize the agent for using “forbidden” info (hide it)
This enables fast learning of policies that respect information constraints.
Hence: Progressive Hiding. 4/6
May 23, 2025 at 1:31 PM
🧠 The idea:
1️⃣ Train as if all information is visible
2️⃣ Penalize the agent for using “forbidden” info (hide it)
This enables fast learning of policies that respect information constraints.
Hence: Progressive Hiding. 4/6
1️⃣ Train as if all information is visible
2️⃣ Penalize the agent for using “forbidden” info (hide it)
This enables fast learning of policies that respect information constraints.
Hence: Progressive Hiding. 4/6
If you've played cooperative imperfect-information games like Hanabi or The Crew, you know how hard they are.
For an AI with no inter-game communication and no perfect recall? Even harder!
That’s where Progressive Hiding comes in. 🔍 3/6
For an AI with no inter-game communication and no perfect recall? Even harder!
That’s where Progressive Hiding comes in. 🔍 3/6
May 23, 2025 at 1:30 PM
If you've played cooperative imperfect-information games like Hanabi or The Crew, you know how hard they are.
For an AI with no inter-game communication and no perfect recall? Even harder!
That’s where Progressive Hiding comes in. 🔍 3/6
For an AI with no inter-game communication and no perfect recall? Even harder!
That’s where Progressive Hiding comes in. 🔍 3/6
Inspired by the classic Progressive Hedging algorithm by Rockafellar & Wets (1991), our method brings similar ideas to multi-agent learning.
That paper has 2000+ citations — and we’re excited to adapt it to game theory!
#AAMAS #AI #GameTheory 2/6
That paper has 2000+ citations — and we’re excited to adapt it to game theory!
#AAMAS #AI #GameTheory 2/6
May 23, 2025 at 1:29 PM
Inspired by the classic Progressive Hedging algorithm by Rockafellar & Wets (1991), our method brings similar ideas to multi-agent learning.
That paper has 2000+ citations — and we’re excited to adapt it to game theory!
#AAMAS #AI #GameTheory 2/6
That paper has 2000+ citations — and we’re excited to adapt it to game theory!
#AAMAS #AI #GameTheory 2/6
Hope everything went well. Thank you Marc!!!!!!!!!!
May 21, 2025 at 11:30 PM
Hope everything went well. Thank you Marc!!!!!!!!!!
Reposted by Benjamin Heymann
The first, at GAIW at 9:55 is "Adaptive Preference Aggregation" by Benjamin Heymann.
This paper proposes a way to adapt a recent urn process algorithm that implements maximal lotteries (by Brandl and @felix-brandt.bsky.social '24) to the setting of user-conditional alignment/recommendation. 2/N
This paper proposes a way to adapt a recent urn process algorithm that implements maximal lotteries (by Brandl and @felix-brandt.bsky.social '24) to the setting of user-conditional alignment/recommendation. 2/N
May 20, 2025 at 1:32 PM
The first, at GAIW at 9:55 is "Adaptive Preference Aggregation" by Benjamin Heymann.
This paper proposes a way to adapt a recent urn process algorithm that implements maximal lotteries (by Brandl and @felix-brandt.bsky.social '24) to the setting of user-conditional alignment/recommendation. 2/N
This paper proposes a way to adapt a recent urn process algorithm that implements maximal lotteries (by Brandl and @felix-brandt.bsky.social '24) to the setting of user-conditional alignment/recommendation. 2/N