George Cai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
George Cai has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
Samuel J Gershman
Harvard University
2 shared publications- Wolfgang Kelsch
Department of Psychiatry and Psychotherapy, University Medical Center, Johannes Gutenberg University Mainz
2 shared publications Albert K Lee
Beth Israel Deaconess Medical Center
1 shared publicationBrian Lustig
Howard Hughes Medical Institute
1 shared publicationConstantine Tsibouris
Department of Molecular and Cellular Biology Harvard University 02138 Cambridge MA USA
1 shared publication- Farhad Pashakhanloo
Center for Brain Science Harvard University 02138 Cambridge MA USA
1 shared publication
Latest Publications
- Open Access
Generalized hippocampal replay exhibits memory recall-like dynamics
Hippocampal replay has been intensely studied for theorized roles in memory consolidation, credit assignment in reinforcement learning, and online planning. Much work in the field assumes that replayed trajectories traverse states smoothly and sequentially, in either forward or reverse order of experience. In a synthetic setting, we demonstrate that incorrect transition priors lead to systematic decoding error and show how model comparison addresses this issue. We compare a diverse set of transition models, including recall-like transition dynamics, and show how they can capture qualitatively different replay structures. Applying model comparison to three neural recording datasets from adult male rats, we find that the recall transition dynamics offer a statistically superior description of spiking during population burst events compared to a commonly used heuristic model of smooth trajectories. We propose a generalized conception of replay events through this theoretical connection to human free recall dynamics. Furthermore, our theory suggests that generalized replays may be a neural instantiation of online planning through generalized rollouts. We find preliminary evidence that generalized replays are more prevalent in settings where planning is beneficial. Overall, the proposed theory has implications for decoding methodology and the normative interpretation of hippocampal replay.
Open AccessGeneralized hippocampal replay exhibits memory recall-like dynamics
Hippocampal replay has been intensely studied for theorized roles in memory consolidation, credit assignment in reinforcement learning, and online planning. Much work in the field assumes that replayed trajectories traverse states smoothly and sequentially, in either forward or reverse order of experience. In a synthetic setting, we demonstrate that incorrect transition priors lead to systematic decoding error and show how model comparison addresses this issue. We compare a diverse set of transition models, including recall-like transition dynamics, and show how they can capture qualitatively different replay structures. Applying model comparison to three neural recording datasets from adult male rats, we find that the recall transition dynamics offer a statistically superior description of spiking during population burst events compared to a commonly used heuristic model of smooth trajectories. We propose a generalized conception of replay events through this theoretical connection to human free recall dynamics. Furthermore, our theory suggests that generalized replays may be a neural instantiation of online planning through generalized rollouts. We find preliminary evidence that generalized replays are more prevalent in settings where planning is beneficial. Overall, the proposed theory has implications for decoding methodology and the normative interpretation of hippocampal replay.
- Open Access
- 18 September 2026
- Published by eLife Sciences Publications, Ltd in eLife preprints
- Vol. 15
Error-driven representation learning in the mesolimbic system
In reinforcement learning, an agent learns to map representations of the environment state to predictions of future reward. Most prior work in neuroscience has assumed a fixed representation and studied how reward prediction errors (thought to be conveyed by phasic dopamine signals) are used to update the mapping from representations to predictions. However, work in machine learning has demonstrated that much more powerful predictive systems can be learned by using the errors to update the representations themselves. We study whether the brain does something similar by leveraging simultaneous recordings of striatal projection neurons in the olfactory tubercle (putatively representing state features) and dopamine neurons in the ventral tegmental area. We show that trial-by-trial changes in striatal activity are more consistent with dopamine-driven representation learning than a variety of alternative updating schemes. This result suggests a convergence of representation learning principles in biological and artificial systems.
- 18 September 2026
- Published by eLife Sciences Publications, Ltd in eLife preprints
- Vol. 15
Open AccessError-driven representation learning in the mesolimbic system
In reinforcement learning, an agent learns to map representations of the environment state to predictions of future reward. Most prior work in neuroscience has assumed a fixed representation and studied how reward prediction errors (thought to be conveyed by phasic dopamine signals) are used to update the mapping from representations to predictions. However, work in machine learning has demonstrated that much more powerful predictive systems can be learned by using the errors to update the representations themselves. We study whether the brain does something similar by leveraging simultaneous recordings of striatal projection neurons in the olfactory tubercle (putatively representing state features) and dopamine neurons in the ventral tegmental area. We show that trial-by-trial changes in striatal activity are more consistent with dopamine-driven representation learning than a variety of alternative updating schemes. This result suggests a convergence of representation learning principles in biological and artificial systems.
- Open Access
Error-driven representation learning in the mesolimbic system
In reinforcement learning, an agent learns to map representations of the environment state to predictions of future reward. Most prior work in neuroscience has assumed a fixed representation and studied how reward prediction errors (thought to be conveyed by phasic dopamine signals) are used to update the mapping from representations to predictions. However, work in machine learning has demonstrated that much more powerful predictive systems can be learned by using the errors to update the representations themselves. We study whether the brain does something similar by leveraging simultaneous recordings of striatal projection neurons in the olfactory tubercle (putatively representing state features) and dopamine neurons in the ventral tegmental area. We show that trial-by-trial changes in striatal activity are more consistent with dopamine-driven representation learning than a variety of alternative updating schemes. This result suggests a convergence of representation learning principles in biological and artificial systems.
Open AccessError-driven representation learning in the mesolimbic system
In reinforcement learning, an agent learns to map representations of the environment state to predictions of future reward. Most prior work in neuroscience has assumed a fixed representation and studied how reward prediction errors (thought to be conveyed by phasic dopamine signals) are used to update the mapping from representations to predictions. However, work in machine learning has demonstrated that much more powerful predictive systems can be learned by using the errors to update the representations themselves. We study whether the brain does something similar by leveraging simultaneous recordings of striatal projection neurons in the olfactory tubercle (putatively representing state features) and dopamine neurons in the ventral tegmental area. We show that trial-by-trial changes in striatal activity are more consistent with dopamine-driven representation learning than a variety of alternative updating schemes. This result suggests a convergence of representation learning principles in biological and artificial systems.
- Open Access
Sparse input representations explain odor discrimination in complex, concentration-varying mixtures
SUMMARY: In natural environments, animals must detect behaviorally relevant odors despite variability in both odor mixture composition and stimulus intensity. Although mice can identify salient odors embedded in complex mixtures, how target concentration and background complexity jointly constrain discrimination remains unclear. We trained mice in a two-alternative forced choice task to identify target odors embedded in mixtures containing up to 16 background components. After performance stabilized, we systematically varied target odor concentration. Discrimination accuracy declined with decreasing target concentration but showed little additional dependence on background complexity. Using a biophysically grounded model of olfactory bulb glomerular responses, we show that linear decoding reproduces behavioral performance when intrinsic neural noise dominates over background-driven variability. Manifold capacity analysis revealed that neural representations remain efficiently structured for odor discrimination despite variation in background complexity. These results define a noise-limited regime of olfactory discrimination in which target detectability is primarily constrained by neural sensitivity rather than background interference.
Open AccessSparse input representations explain odor discrimination in complex, concentration-varying mixtures
SUMMARY: In natural environments, animals must detect behaviorally relevant odors despite variability in both odor mixture composition and stimulus intensity. Although mice can identify salient odors embedded in complex mixtures, how target concentration and background complexity jointly constrain discrimination remains unclear. We trained mice in a two-alternative forced choice task to identify target odors embedded in mixtures containing up to 16 background components. After performance stabilized, we systematically varied target odor concentration. Discrimination accuracy declined with decreasing target concentration but showed little additional dependence on background complexity. Using a biophysically grounded model of olfactory bulb glomerular responses, we show that linear decoding reproduces behavioral performance when intrinsic neural noise dominates over background-driven variability. Manifold capacity analysis revealed that neural representations remain efficiently structured for odor discrimination despite variation in background complexity. These results define a noise-limited regime of olfactory discrimination in which target detectability is primarily constrained by neural sensitivity rather than background interference.