Xiao-Xiong Lin
@xxlin.bsky.social
Computational and systems neuroscience. DeepRL Hippocampus navigation <- Data analysis / neural network modelling PFC working memory flexible cognition.
xiaoxionglin.com
https://www.bcf.uni-freiburg.de/about/people/lin
github.com/xiaoxionglin/dSCA
xiaoxionglin.com
https://www.bcf.uni-freiburg.de/about/people/lin
github.com/xiaoxionglin/dSCA
Completely agree🫡 We’ve got a freshly baked contribution along this path:
bsky.app/profile/xxli...
bsky.app/profile/xxli...
🎉 Accepted at ICLR 2026! 🎉
We show that place-cell–like spatial representations can emerge in a deep RL agent with structured recurrent dynamics (like hippocampus🌊🐴), without explicit spatial supervision.
PDF: openreview.net/forum?id=li1...
We show that place-cell–like spatial representations can emerge in a deep RL agent with structured recurrent dynamics (like hippocampus🌊🐴), without explicit spatial supervision.
PDF: openreview.net/forum?id=li1...
Emergence of Spatial Representation in an Actor-Critic Agent with...
Sequential activation of place-tuned neurons in an animal during navigation is typically interpreted as reflecting the sequence of input from adjacent positions along the trajectory. More recent...
openreview.net
January 28, 2026 at 2:36 AM
Completely agree🫡 We’ve got a freshly baked contribution along this path:
bsky.app/profile/xxli...
bsky.app/profile/xxli...
Would love to hear thoughts from both ML and neuroscience folks on using RL as a functional testbed for brain circuit models🧠🤖🎰
January 28, 2026 at 2:27 AM
Would love to hear thoughts from both ML and neuroscience folks on using RL as a functional testbed for brain circuit models🧠🤖🎰
Takeaway:
Modern deep reinforcement learning provides a principled testbed for hippocampal circuit hypotheses, supporting a view in which intrinsic CA3 sequence dynamics scaffold spatial representations from egocentric experience rather than merely reflecting replay or planning.
8/8
Modern deep reinforcement learning provides a principled testbed for hippocampal circuit hypotheses, supporting a view in which intrinsic CA3 sequence dynamics scaffold spatial representations from egocentric experience rather than merely reflecting replay or planning.
8/8
January 28, 2026 at 2:22 AM
Takeaway:
Modern deep reinforcement learning provides a principled testbed for hippocampal circuit hypotheses, supporting a view in which intrinsic CA3 sequence dynamics scaffold spatial representations from egocentric experience rather than merely reflecting replay or planning.
8/8
Modern deep reinforcement learning provides a principled testbed for hippocampal circuit hypotheses, supporting a view in which intrinsic CA3 sequence dynamics scaffold spatial representations from egocentric experience rather than merely reflecting replay or planning.
8/8
Unlike many prior approaches that explicitly encourage spatial structure (e.g. via mapping or auxiliary losses), our model includes no spatial objectives.
Nevertheless, structured spatial tuning emerges during navigation.
7/n
Nevertheless, structured spatial tuning emerges during navigation.
7/n
January 28, 2026 at 2:22 AM
Unlike many prior approaches that explicitly encourage spatial structure (e.g. via mapping or auxiliary losses), our model includes no spatial objectives.
Nevertheless, structured spatial tuning emerges during navigation.
7/n
Nevertheless, structured spatial tuning emerges during navigation.
7/n
The sequence-based agent develops place-cell–like spatial tuning and distance-dependent representational similarity of spatial locations.
By contrast, LSTM agents trained on the same tasks do not form comparably structured spatial representations.
6/n
By contrast, LSTM agents trained on the same tasks do not form comparably structured spatial representations.
6/n
January 28, 2026 at 2:21 AM
The sequence-based agent develops place-cell–like spatial tuning and distance-dependent representational similarity of spatial locations.
By contrast, LSTM agents trained on the same tasks do not form comparably structured spatial representations.
6/n
By contrast, LSTM agents trained on the same tasks do not form comparably structured spatial representations.
6/n
With sparse sensory encoding, agents using intrinsic sequences learn faster and more stably than standard recurrent baselines, including LSTM agents.
This advantage largely disappears when input is dense.
5/n
This advantage largely disappears when input is dense.
5/n
January 28, 2026 at 2:18 AM
With sparse sensory encoding, agents using intrinsic sequences learn faster and more stably than standard recurrent baselines, including LSTM agents.
This advantage largely disappears when input is dense.
5/n
This advantage largely disappears when input is dense.
5/n
We embed a minimal, interpretable DG–CA3–like sequence generator core in an end-to-end actor–critic agent operating in a realistic navigation environment, without auxiliary objectives.
4/n
4/n
January 28, 2026 at 2:14 AM
We embed a minimal, interpretable DG–CA3–like sequence generator core in an end-to-end actor–critic agent operating in a realistic navigation environment, without auxiliary objectives.
4/n
4/n
At a mechanistic level, hippocampal circuits generate intrinsic activity sequences even when sequential sensory input is sparse or absent.
This suggests intrinsic sequence dynamics as a plausible substrate for constructing spatial representations from egocentric experience.
3/n
This suggests intrinsic sequence dynamics as a plausible substrate for constructing spatial representations from egocentric experience.
3/n
January 28, 2026 at 2:04 AM
At a mechanistic level, hippocampal circuits generate intrinsic activity sequences even when sequential sensory input is sparse or absent.
This suggests intrinsic sequence dynamics as a plausible substrate for constructing spatial representations from egocentric experience.
3/n
This suggests intrinsic sequence dynamics as a plausible substrate for constructing spatial representations from egocentric experience.
3/n
In real-world navigation, the sensory stream is ambiguous and policy-dependent, while spatially informative “landmarks" are sparse.
Many biological models emphasize interpretability but lack task-level realism, while engineering approaches achieve competence with limited mechanistic insight.
2/n
Many biological models emphasize interpretability but lack task-level realism, while engineering approaches achieve competence with limited mechanistic insight.
2/n
January 28, 2026 at 2:01 AM
In real-world navigation, the sensory stream is ambiguous and policy-dependent, while spatially informative “landmarks" are sparse.
Many biological models emphasize interpretability but lack task-level realism, while engineering approaches achieve competence with limited mechanistic insight.
2/n
Many biological models emphasize interpretability but lack task-level realism, while engineering approaches achieve competence with limited mechanistic insight.
2/n