arXiv subjects

A. Lorincz

Publications and source records attributed to A. Lorincz.

6 recordsLinked to original sources

arXiv2006

Mind model seems necessary for the emergence of communication

We consider communication when there is no agreement about symbols and meanings. We treat it within the framework of reinforcement learning. We apply different reinforcement learning models in our studies and simplify the problem as much as possible. We show that the modelling of the other agent is insufficient in the simplest possible case, unless the intentions can also be modelled. The model of the agent and its intentions enable quick agreements about symbol-meaning association. We show that when both agents assume an `intention model' about the other agent then the symbol-meaning association process can be spoiled and symbol meaning association may become hard.

q-bio.NC↗

arXiv2005

Selection in Scale-Free Small World

In this paper we compare the performance characteristics of our selection based learning algorithm for Web crawlers with the characteristics of the reinforcement learning algorithm. The task of the crawlers is to find new information on the Web. The selection algorithm, called weblog update, modifies the starting URL lists of our crawlers based on the found URLs containing new information. The reinforcement learning algorithm modifies the URL orderings of the crawlers based on the received reinforcements for submitted documents. We performed simulations based on data collected from the Web. The collected portion of the Web is typical and exhibits scale-free small world (SFSW) structure. We have found that on this SFSW, the weblog update algorithm performs better than the reinforcement learning algorithm. It finds the new information faster than the reinforcement learning algorithm and has better new information/all submitted documents ratio. We believe that the advantages of the selection algorithm over reinforcement learning algorithm is due to the small world property of the Web.

cs.LG↗

arXiv2004

L1 regularization is better than L2 for learning and predicting chaotic systems

Emergent behaviors are in the focus of recent research interest. It is then of considerable importance to investigate what optimizations suit the learning and prediction of chaotic systems, the putative candidates for emergence. We have compared L1 and L2 regularizations on predicting chaotic time series using linear recurrent neural networks. The internal representation and the weights of the networks were optimized in a unifying framework. Computational tests on different problems indicate considerable advantages for the L1 regularization: It had considerably better learning time and better interpolating capabilities. We shall argue that optimization viewed as a maximum likelihood estimation justifies our results, because L1 regularization fits heavy-tailed distributions -- an apparently general feature of emergent systems -- better.

cs.LG↗

arXiv2004

Applying Policy Iteration for Training Recurrent Neural Networks

Recurrent neural networks are often used for learning time-series data. Based on a few assumptions we model this learning task as a minimization problem of a nonlinear least-squares cost function. The special structure of the cost function allows us to build a connection to reinforcement learning. We exploit this connection and derive a convergent, policy iteration-based algorithm. Furthermore, we argue that RNN training can be fit naturally into the reinforcement learning framework.

cs.AI↗

arXiv2003

Comparation based bottom-up and top-down filtering model of the hippocampus and its environment

Two rate code models -- a reconstruction network model and a control model -- of the hippocampal-entorhinal loop are merged. The hippocampal-entorhinal loop plays a double role in the unified model, it is part of a reconstruction network and a controller, too. This double role turns the bottom-up information flow into top-down control like signals. The role of bottom-up filtering is information maximization, noise filtering, temporal integration and prediction, whereas the role of top-down filtering is emphasizing, i.e., highlighting or `paving of the way' as well as context based pattern completion. In the joined model, the control task is performed by cortical areas, whereas reconstruction networks can be found between cortical areas. While the controller is highly non-linear, the reconstruction network is an almost linear architecture, which is optimized for noise estimation and noise filtering. A conjecture of the reconstruction network model -- that the long-term memory of the visual stream is the linear feedback connections between neocortical areas -- is reinforced by the joined model. Falsifying predictions are presented; some of them have recent experimental support. Connections to attention and to awareness are made.

q-bio.NC↗

arXiv2002

HebbNets: Dynamic network with Hebbian learning rule

It has been demonstrated that one of the most striking features of the nervous system, the so called 'plasticity' (i.e high adaptability at different structural levels) is primarily based on Hebbian learning which is a collection of slightly different mechanisms that modify the synaptic connections between neurons. The changes depend on neural activity and assign a special dynamic behavior to the neural networks. From a structural point of view, it is an open question what network structures may emerge in such dynamic structures under 'sustained' conditions when input to the system is only noise. In this paper we present and study the `HebbNets', networks with random noise input, in which structural changes are exclusively governed by neurobiologically inspired Hebbian learning rules. We show that Hebbian learning is able to develop a broad range of network structures, including scale-free small-world networks.

nlin.AO↗