Takuma Torii

Platform design for large-scale artificial market simulation and preliminary evaluation on the K computer

Artificial market simulations have the potential to be a strong tool for studying rapid and large market fluctuations and designing financial regulations. High-frequency traders, that exchange multiple assets simultaneously within a millisecond, are said to be a cause of rapid and large market fluctuations. For such a large-scale problem, this paper proposes a software or computing platform for large-scale and high-frequency artificial market simulations (Plham: /pl

Toward a Mechanistic Account for Imitation Learning: An Analysis of Pendulum Swing-Up

\Lambda

Differentiation Through Symbolic Communication

Λm). The computing platform, Plham, enables modeling financial markets composed of various brands of assets and a large number of agents trading on a short timescale. The design feature of Plham is the separation of artificial market models (simulation models) from their execution (execution models). This allows users to define their simulation models without parallel computing expertise and to choose one of the execution models they need. This computing platform provides a prototype execution model for parallel simulations, which exploits the variety in trading frequency among traders, that is, the fact that some traders do not require up-to-date information of markets changing in millisecond order. We evaluated a prototype implementation on the K computer using up to 256 computing nodes.

Blending defined on difference, and an implementation using dynamical approach

Symbolic communication is a dynamic process involving both generation and sharing of novel expressions and meanings. In order to understand the mechanism of the dynamic process, we construct a computational model of communicative agents, which realise the generation and sharing. Our proposed model is a hybrid system of a discrete symbol processing system and a dynamical system processing continuous variables. The former is implemented by a rewriting rule system and the latter by a neural network. We found that the communication processes in this model was classified into three types: non-generative, partially-generative, and generative. In the generative type, the generation of novel utterances and meanings is realised. We show that the novel utterances are generated through linguistic analogy which is to extend application of rewriting rules in individuals knowledge in the symbolic system. We also indicate that the neural network is structured in communication and shows dynamic change of dynamical states, which contributes to the generation of novel meanings.

Shock transfer by arbitrage trading: analysis using multi-asset artificial market

Learning an action from others require to infer their underlying goals, and recent psychological studies have reported behavioral evidences that young children do infer others’ underlying goals by observing their actions. The goal of the present study is to propose a mechanistic account for how this goal inference is possible by observing others’ actions. For this purpose, we performed a series of simulations in which two agents control pendulums toward different goals, and analyzed with which types of features it is possible to infer their different latent goals and control schemes. Our analysis showed that pointwise dimension, a type of fractal dimension, of the pendulum movements is sufficiently informative to classify the types of agents. With respect to its invariant nature, this result suggests that the fine-grained movement patterns such as the fractal dimension reflect the structure of the underlying control schemes and goals.

Which types of learning make a simple game complex

Cooperation is a key to understand social behavior and decision-making in conflict situations in nature or society. The problem of cooperation is formulated as Iterated Prisoners Dilemma in game theory. To find out pairs of strategies that establish mutual cooperation in equilibrium, several studies used evolutionary computer simulations. Although the payoff matrix of IPD is defined by the two ordinal inequalities, using simulations requires numerical payoff values. Since Axelrod used a numerical payoff matrix, many researchers have adopted the same payoff values with no justification. We guess that there is the common assumption that findings obtained by examining one numerical IPD payoff matrix hold in more general cases. However, this is not trivial, because we have no evidence supporting the assumption. In this article, we verify this assumption by analyzing IPD games of two payoff-maximizing players with memory-1 Markov strategies. To determine equilibria of IPD, we define an extension of Nash equilibrium with the partial derivatives of payoff functions. Our numerical and formal analyses falsify the assumption. We showed the formal evidence for the conditions that the WSLS pair does not become an equilibrium for some IPD games. Contrarily, our results suggest that the TFT pair might be an equilibrium for any IPD games. Falsification of the common assumption requests a new classification of strategy pairs as equilibria in IPD.

Impact of Market Maker to Efficiency of Stock Market with Dark Pool - Artificial Market Approach

We explore the manner in which a neuro-symbolic hybrid system differentiates through symbolic communication. The hybrid system consists of a recurrent neural network and a rewriting rule system. It is shown that the differentiation comes from the asymmetric structure of the symbol system, i.e., usage of different rules in deriving and accepting the same symbolic message. The dynamics of symbolic communication depends on the phase space structure of the neural system. The asymmetric structure is realized by generalization, and it must be ubiquitous in adaptive symbol systems such as human language.

Toward a mechanistic account for imitation learning: an analysis of pendulum swing-up

The human cognitive ability to blend mental objects (blending) has been a subject of interest in cognitive linguistics. Some researchers are motivated to give formal descriptions to concepts developed in cognitive linguistics. We have formulated conditions of blending that provide testable conditions to classify whether the final product of blending could be uniquely produced as the result of processing the whole of the word combinations. Our formulation allows us to theoretically test the conditions of blending up to an indefinitely large series of word combinations. Such a formal description of the conditions of blending has not appeared in previous studies. We demonstrated our formulation by implementing a neural network model, as a dynamical system. As the results of computer simulations showed, we found that the model exhibits “context-dependency” and “ambiguity” (probabilistic behavior) in the process of blending. This can be seen as a model of communication, more generally, demonstrating a process that involves a series of making of meanings.