Matthew Riemer

[Matthew Riemer @MattRiemer]

7/7: Grateful to my co-authors @Tommaso_Tosato , Amin Memarian, @mptouzel, @GlenBerseth, @irinarish, and @introspection.

A collaboration across @Mila_Quebec, @UMontreal, @IBMResearch, and @TaraResearch, supported by @foresightinst.

[Matthew Riemer @MattRiemer]

1/7: 🚨 AI agents are predisposed to collude. 🤖📉

Our #ICML2026 paper shows that reasoning models (like DeepSeek-R1) tacitly collude on prices with competing firms, even when explicitly told NOT to.

And their chain-of-thought reveals nothing. 🧵

[Matthew Riemer @MattRiemer]

6/7: 📖 Paper: https://t.co/l3RZb2CGJu
💻 Code: https://t.co/t3JoqkmvF1

At #ICML in Seoul? Come by poster #3309 at 5pm!

[Glen Berseth @GlenBerseth]

#Reinforcementlearning and regret minimization are not designed for the real world, where it takes time to compute an action. In this work, we prove bounds on real-time regret and show how algorithms based on these principles reduce wall-clock regret. #regretinTIMEnotactions

[Rishav @rishav_real]

Here is the link to our poster session: https://t.co/ysEu0DGKiS.

Also, find @gopeshh1 presenting a orthogonal but closely related work on real-time RL. Both works complement each other and we will soon release a comprehensive blog talking about it!

[Gopeshh Subbaraj @gopeshh1]

1/
 Most RL methods assumes a turn-based setup-- agent acts, environment responds. But in the real world, the environment doesn’t wait.
In real-time RL, slow inference means missed actions or delayed ones. This leads to two key challenges:
 • Inaction Regret
 • Delay Regret

[Gopeshh Subbaraj @gopeshh1]

Do drop by poster #609 in Hall 2 to hear more about this work. #ICLR2025
@Mila_Quebec

[Rishav @rishav_real]

Find our work @iclr_conf, presented by @plaisir_avec (Hall 3, Poster #428) and done @Mila_Quebec! We introduce pipelining + temporal skip connections to tackle delay in real-time RL-speeding up inference, reducing depth-wise delay, and restoring stability in large models.

[Germán Abrevaya @GAbrevaya]

Thrilled to announce our paper on integrating domain knowledge into machine learning models & encoding whole-brain dynamics into low-dimensional dynamical systems is now published at @TmlrOrg! 🎉

https://t.co/YYPCDaLouY
https://t.co/40Mws9WGTB

#SciML #TimeSeries #Neuroscience

[J. AI Research-JAIR @JAIR_Editor]

New Article: "Towards Continual Reinforcement Learning: A Review and Perspectives" by Khetarpal, Riemer, Rish and Precup https://t.co/7ZYBG5IOcL

[Massimo Caccia @MassCaccia]

The RL track for our CVPR Continual Learning challenge is out 😱 Submissions are opened till May 23rd. And don't forget, the biggest prize is given to the team that performs the best in both the SL and RL track 🤑

[Massimo Caccia @MassCaccia]

Our CVPR Continual Learning challenge is out 🚀
You can try the supervised learning track right now and the RL one should come anytime soon!
There's a lot of reason why you should try it out... 🧵time 👇

Challenge: https://t.co/agNP7vz7A7

Sequoia: https://t.co/lwHt3gkkKV

1/N

[Vincenzo Lomonaco @v_lomonaco]

🔥 Tomorrow, 17.30 CET, the @ContinualAI reading group 🔥

You cannot miss it if you are into Continual Learning & Coding! :)
@MassCaccia and Fabrice Normandin from @Mila_Quebec will talk about their 4recent work "Sequoia - The Research Tree"!

https://t.co/oqlA5U8lcs

[Matthew Riemer @MattRiemer]

Check out our new blog post explaining an algorithm for deep continual learning presented at ICLR this morning called Meta-Experience Replay! https://t.co/okhXlXCuZh #ICLR2019

[Futurism @futurism]

AI systems can now adapt to their changing environments. https://t.co/d63RKYZnwt

[ZDNET @ZDNET]

IBM’s AI flies back and forth through time in Flappy Bird https://t.co/3z8fcphWXD