\documentclass{seminar} \usepackage{colordvi} \begin{document}
The hippocampus and temporal sequence learning
a review of the primary literature
\begin{center}\emph{Cite and summarize the primary literature on the role of the hippocampus in the storage and retrieval of sequence memories (learning of a temporal sequence of stimuli)}\end{center}
Hippocampal lesions
---
\RawSienna{$\to$ problems remembering sequences}
- Sequences of spatial locations: Chiba et al '94
- Odor sequences: Kesner et al '02, Fortin et al '02
---
\RawSienna{$\to$ problems keeping track of sequences as they occur}
- Agster et al '02
- imaging study shows activation: Takakura et al '03 \ \ \ \ \ \ \begin{scriptsize}alternate interpretation: keeping track of context, in general\end{scriptsize}
A closer look at Fortin et al '02
\includegraphics[scale=.5]{fortin1.eps}
- lesions $\Rightarrow$
- \ \ \ given two items, can't remember \RawSienna?{which is earlier}
- \ \ \ ...but can still tell \ForestGreen?{if an item was in the sequence at all}
- $\to$ \textbf{only a single dissociation} $\gets$
Wouldn't it be nice if we could look inside the hippocampus after learning and see if any memories of sequences are in there?
Representations:
Reply of sequences found during sleep
- Lee and Wilson '02 (SWS)
- Louie and Wilson '01 (REM)
- Kudrimoti et al '99 (SWS)
- Qin et al '97 (SWS)
- Skaggs and McNaughton? '96 (SWS)
Hippocampus is not the \underline{only} structure involved with sequences
---
\Mahogany{Serial Reaction Time Task (SRTT)}:
implicit sequence learning w/o hippocampus
- hippocampus is not necc. for SRTT \begin{scriptsize}(Christie and Dalrymple-Alford '04)\end{scriptsize}
- other studies show that it might be helpful, though \begin{scriptsize}(Curran '97)\end{scriptsize}, and an imaging study detects activation \begin{scriptsize}(Schendan et al '03)\end{scriptsize}
Other structures that may learn sequences:
- basal ganglia
- cerebellum
- prefrontal cortex
Generalizations of temporal sequence learning
Evidence of hippocampal involvement in sequence memory
$\nRightarrow$
there is machinery specialized to sequence memory in the hippocampus
- \textbf{maybe it contains a subroutine used by sequence learning}
- \textbf{maybe it contains a generalization of sequence learning}
Generalizations of temporal sequence learning cont'd
- \textbf{Episodic memory}
- \textbf{Declarative/explicit memory}
- \textbf{Relational inference}
- \textbf{Temporal reasoning}
Generalizations of temporal sequence learning cont'd
- \textbf{Episodic memory}
- \textbf{Declarative/explicit memory}
- \textbf{Relational inference} \ \ \ \ \emph{transitive inference}: $A < B$, $B < C$ \Rightarrow $A < C$
- \ \ \ \ \textbf{The task in Fortin et al '02 is a special case of transitive inference}
- \ \ \ \ \begin{scriptsize}(Dusek and Eichenbaum '97)\end{scriptsize} said that hippocampus neccessary for transitive inference, but \begin{scriptsize}(Van Elzakker et al '03)\end{scriptsize} showed that study was probably flawed
- \ \ \ \ imaging studies show hippocampal activation during transitive inference \begin{scriptsize}(Heckers et al '04) (Preston et al '04) (Nagode and Pardo '02)\end{scriptsize}, but see \begin{scriptsize}(Acuna et al '02)\end{scriptsize} \begin{tiny}((only during familiar situations??))\end{tiny}
Generalizations of temporal sequence learning cont'd
- \textbf{Episodic memory}
- \textbf{Declarative/explicit memory}
- \textbf{Relational inference}
- \textbf{Temporal reasoning}
- \ \ \ \ Hippocampus necc. to distinguish between 2-second and 8-second interval \begin{scriptsize}(Jackson et al '98)\end{scriptsize}
- \ \ \ \ Damage impairs memory of temporal distance in humans \begin{scriptsize}(Hopkins et al '95)\end{scriptsize}
- \ \ \ \ Damage impairs trace conditioning \begin{scriptsize}(Clark and Squire '98)\end{scriptsize} \begin{scriptsize}(Moyer et al '90)\end{scriptsize} \begin{scriptsize}(Woodruff-Pak '93)\end{scriptsize} \begin{scriptsize}(Beylin et al '01)\end{scriptsize}
Computational models of hippocampal sequence learning
- Theta multiplexed model
- Levy '96: A sequence predicting CA3 is a flexible associator that learns and uses context to solve hippocampal-like tasks.
- Hasselmo-Wallenstein Model
- Sato N, Yamaguchi Y '03: Memory encoding by "Theta Phase Precession" in the hippocampal network
- Tsodyks et al '96
The theta multiplexed model
\includegraphics[scale=.7]{thetaModel1.eps}
% ==== The theta multiplexed model cont'd ====
Brain regions involved in working memory oscillate at two rates at once; a low-frequency theta oscillation (5-8 Hz) is subdivided into about seven subcycles by high frequency gamma oscillations (20-60 Hz). Perhaps this is a multiplexed short-term memory buffer; each of the seven subcycles holds one memory. So, during each theta cycle, the brain runs through each of the stored working memories.
\includegraphics[scale=.55]{thetaModel1.eps}
The theta multiplexed model cont'd
\includegraphics[scale=.7]{thetaModelSchematic.eps}
The theta multiplexed model cont'd
Sequence readout with a synchronized decoder wave: phase determines which sequence location is read:
Element A selected
\includegraphics[scale=.6]{thetaSyncReadout.eps}
Element C selected
\includegraphics[scale=.6]{thetaSyncReadout2.eps}
The theta multiplexed model cont'd
Element C selected
\includegraphics[scale=.6]{thetaSyncReadout2.eps}
No element selected
(decoder wave in antiphase with carrier wave)
\includegraphics[scale=.6]{thetaSyncReadoutNone.eps}
The theta multiplexed model cont'd
Sequence readout with a slightly slower decoder wave (using precession):
\includegraphics[scale=.7]{thetaNormalReadout.eps}
The theta multiplexed model cont'd
\RawSienna?{\textbf{Explains theta phase precession in hippocampus:}}
\includegraphics[scale=1]{thetaPhasePrecession.eps}
The wave shown in the decoder wave. The rat is replying a sequence of nearby places into it's short-term memory. At first, as it is at location A, the "C" place cell is activated (because it's nearby C). Perhaps it remembers sequence A,B,C,D,E. Then, it moves to location B. The "C" place cell now replays sequence "BCDE A". etc
More about theta oscillation
- Characteristic oscillation (or rhythm) in the hippocampus: 4-12 hz.
- Seems to be imposed on the hippocampus by the medial septum-diagonal band of Broca \begin{scriptsize}(Buzsaki '02)\end{scriptsize}.
- Seen during exploratory activity and REM sleep.
- Can be evoked and its phase can be reset by the presentation of sensory stimuli \begin{scriptsize}(Tesche and Karhu '97)\end{scriptsize}.
More about theta phase precession
Theta phase precession is a phenomenon relating the theta rhythm, the firing of individual neurons, and place fields. Consider the phase of each spike of the place cell in relation to the hippocampal theta oscillation. Theta phase precession is a phenomenon where the phase of neural firing moves earlier each theta cycle as the animal moves through the place field.
Some postulated functions of theta rhythm and theta phase precession:
- holding many memories in mind at once
- rehersing nearby events in a remembered sequence
- a "program counter" which determines which step of a learning algorithm is currently being executed.
- a carrier frequency for neural coding based on spike phase
The theta multiplexed model cont'd
Theta phase coding lets you do other cool stuff, too. For example, sequence comparison during readout (no extra "compare" op needed):
\includegraphics[scale=1]{thetaCompare.eps}
The theta multiplexed model cont'd
Provides some intriguing answers to some questions:
\RawSienna?{Q: Why does working memory hold 7 +- 2 items?}
A: There's room for about seven gamma cycles inside each theta cycle.
The theta multiplexed model cont'd
\RawSienna?{Q: Why is possibly transient frequency-locking observed between the hippocampus and the entorhinal cortex, the cingulate cortex, and the prefrontal cortex?}
A: Phase coding means transmitting information in the phase of individual spikes, as compared to an underlying oscillation. If there is a phase-coded signal being transferred between the hippocampus and one of these other areas, then both sender and receiver must be frequency-locked in order to be able to compare the phases of the spikes to the same underlying oscillation.
\end{document}
Bibliography