連想記憶に最適な神経実装の探索: 海馬のスパイクタイミング依存可塑性と位相応答曲線
| Autor: | Miyata, Ryouta |
|---|---|
| Jazyk: | japonština |
| Rok vydání: | 2014 |
| Předmět: | |
| Popis: | 近年、記憶を記銘するためのシナプス可塑性則と想起のための神経相互作用則の最適な組合せを探索する試みがある。Lengyelら（2005）は、スパイクタイミング依存可塑性(spike-timing-dependent plasticity, STDP) に着目した場合、その記憶想起には位相応答曲線（phase response curve, PRC）で記述された神経相互作用を用いることが適切であると仮定し、ベイズ理論の枠組みから最適なSTDPとPRCの関係を導いた。本研究では、神経細胞の高次元非線形動力学系の位相縮約モデルを求め、このモデルの物理的制約下で最適な記銘と想起を実現するSTDPとPRCの組合せを導出し、Lengyelらとの対応を明らかにした。また、最近の研究から、海馬ではスパイク時空間パターンの記銘とその時間反転パターンや引き延ばしパターンの想起が行われていることが示唆されている。そこで本手法をこれらの記憶想起の場合に適用し、最適なSTDPとPRCの組合せを探索した。 Recently, some researchers have tried to specify optimal pairs of the synaptic plasticity rule for memory storage and the form of neural interactions for memory retrieval in a neural network model. Lengyel et al. developed a normative theory for auto-associative memory networks recalling spatio-temporal spike patterns on the basis of Bayesian theory. Under the speculation that a phase response curve (PRC) is an appropriate way to formulate the neural interactions if memories are stored by a spike-timing-dependent plasticity (STDP) rule, they derived the optimal pairs of STDP window functions and PRCs for executing the associative memory. In this doctoral thesis, we propose a synthetic approach of optimal design for associative memory networks, which consists of bottom-up and top-down steps: In the bottom-up step, under the assumption of regular spiking and weak coupling, we formulate an associative memory network recalling spatio-temporal patterns as a phase oscillator model consisting of an STDP window function and a PRC. Next, we analytically derive the mutual information between a stored pattern and the retrieval one, and use it to evaluate memory retrieval performance. In the top-down steps, by maximizing the objective function given by the mutual information, we derive optimal pairs of STDP window functions and PRCs. We firstly verify whether the optimal pair of STDP window functions and PRCs for executing the auto-associative memory derived from our approach are identical those derived from Lengyel's approach. With a zero noise limit, we obtain the same relation between the STDP window function and the PRC as that which Lengyel et al. obtained. Furthermore, recently reported experimental findings suggest that the hippocampal CA1 network stores spatio-temporal spike patterns and retrieves temporally reversed and spread-out patterns. We secondly applied to our approach to derive optimal pairs of STDP window functions and PRCs for executing such memory retrievals. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|