Excited to share our new paper accepted at ICML 2026 with @tyrell_turing and Doina Precup!
π°π· See you in Seoul.
A major challenge in continual reinforcement learning is balancing:
β’ plasticity (learning new things)
β’ stability (not forgetting old ones)
π§΅
1/15
Excited to share our new paper accepted at ICML 2026 with @tyrell_turing and Doina Precup!
π°π· See you in Seoul.
A major challenge in continual reinforcement learning is balancing:
β’ plasticity (learning new things)
β’ stability (not forgetting old ones)
π§΅
1/15
@AndrewLampinen We found that under gradual non-stationarity, severe losses of plasticity were indeed much less apparent than in many task-switching settings. However, continual learning remained challenging because stability, and not plasticity, became the primary bottleneck.
Also, special thanks to @ckaplanis1 for sharing his expertise on synaptic consolidation throughout this project. Many of the ideas in this work benefited greatly from our discussions.
Excited to share our new paper accepted at ICML 2026 with @tyrell_turing and Doina Precup!
π°π· See you in Seoul.
A major challenge in continual reinforcement learning is balancing:
β’ plasticity (learning new things)
β’ stability (not forgetting old ones)
π§΅
1/15
Paper: https://t.co/WaDGp7edj9
Blog: https://t.co/1YTUel0gb3
Code:
MuJoCo / PyTorch: https://t.co/QGEhp25GPG
Four Rooms / Jax: https://t.co/b5JktzijOS
Looking forward to discussions at ICML 2026 in Seoul π°π·
15/15
π’ Applications for the Kempner AI Fellows Program are now open!
This program is designed for early-career researchers with strong technical preparation in modern machine learning. Applications close June 1st.
Learn more and apply π https://t.co/GtzKq9Yhpo
#KempnerInstitute