https://www2.econ.iastate.edu/tesfatsi/DeepLearningInNeuralNetworksOverview.JSchmidhuber2015.pdf

Deep learning in neural networks

▪ Abstract The prefrontal cortex has long been suspected to play an important role in cognitive control, in the ability to orchestrate thought and action in accordance with internal goals. Its neural basis, however, has remained a mystery. Here, we propose that cognitive control stems from the active maintenance of patterns of activity in the prefrontal cortex that represent goals and the means to achieve them. They provide bias signals to other brain structures whose net effect is to guide the flow of activity along neural pathways that establish the proper mappings between inputs, internal states, and outputs needed to perform a given task. We review neurophysiological, neurobiological, neuroimaging, and computational studies that support this theory and discuss its implications as well as further issues to be addressed

/pdf/an-integrative-theory-of-prefrontal-cortex-function-2j9unprn5f.pdf

An integrative theory of prefrontal cortex function

http://www.maths.qmul.ac.uk/%7Elatora/report_06.pdf

Complex networks: Structure and dynamics

Damage to the hippocampal system disrupts recent memory but leaves remote memory intact. The account presented here suggests that memories are first stored via synaptic changes in the hippocampal system, that these changes support reinstatement of recent memories in the neocortex, that neocortical synapses change a little on each reinstatement, and that remote memory is based on accumulated neocortical changes. Models that learn via changes to connections help explain this organization. These models discover the structure in ensembles of items if learning of each item is gradual and interleaved with learning about other items. This suggests that the neocortex learns slowly to discover the structure in ensembles of experiences. The hippocampal system permits rapid learning of new items without disrupting this structure, and reinstatement of new memories interleaves them with others to integrate them into structured neocortical memory systems.

Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory.

The theory of affordances

Deep reinforcement learning (RL) has made it possible to solve complex robotics problems using neural networks as function approximators. However, the policies trained on stationary environments suffer in terms of generalization when transferred from one environment to another. In this work, we use Robust Markov Decision Processes (RMDP) to train the drone control policy, which combines ideas from Robust Control and RL. It opts for pessimistic optimization to handle potential gaps between policy transfer from one environment to another. The trained control policy is tested on the task of quadcopter positional control. RL agents were trained in a MuJoCo simulator. During testing, different environment parameters (unseen during the training) were used to validate the robustness of the trained policy for transfer from one environment to another. The robust policy outperformed the standard agents in these environments, suggesting that the added robustness increases generality and can adapt to non-stationary environments. Codes: https://github.com/adipandas/gym_multirotor

Robust Deep Reinforcement Learning for Quadcopter Control.

‘There’s no art to find the mind’s construction in the face,’ wrote Shakespeare, but trying to infer what someone is really thinking is arguably the essence of interaction between cognitive agents. Turning this into a computational model is challenging, but one possible approach to infer mental models from linguistic expression is to look at patterns of lexical associations. Assuming that written language reflects conceptual associations in the writer’s mind, we have previously shown differences in the patterns of lexical association between creative and non-creative writing. In this paper, we apply the same approach to news reports from individual media sources over the same period, with the goal of looking for differential associative patterns. The underlying assumption is that the associative patterns of a media source will reflect its “mind” and “personality,” i.e., specific styles, preferences, or biases, just as they do for individuals.

Reading the Media’s Mind

Due to its effectiveness, the amount of projects and people involved in research of Intelligent Systems is growing. This positive trend has resulted in the increase in demand for academic educational courses/learning experiences especially at the graduate level. This paper focuses on the approaches of two educators at the University of Cincinnati in the area of fuzzy logic and intelligent & adaptive systems. The lessons learned and the winning strategies are underscored in an attempt to enhance the open exchange of ideas, teaching styles and methodologies.

Teaching Intelligent Systems at the University of Cincinnati

Thinking is a self-organized dynamical process and, as such, interesting to characterize. However, direct, real-time access to thought at the semantic level is still very limited. The best that can be done is to look at spoken or written expression. The question we address in this research is the following: Is there a characteristic pitch of thought? To begin answering this complex question, we look at text documents from several large corpora at the sentence level – i.e., using sentences as the units of meaning – and considering each document to be the result of a random process in semantic space. Given a large corpus of multi-sentence documents, we build a lexical association network representing associations between words in the corpus. This network is used to induce a semantic similarity metric between sentences, and each document is segmented into multi-sentence semantically coherent blocks (SCBs) with occasional connecting text between the blocks. Based on this segmentation, the process of document generation is modeled as a sticky Markov chain at the sentence level. We show that most documents across all the corpora are sequences of blocks with a very consistent mean length of 6.4 sentences across the corpora. This consistency suggests that a value of 6-7 sentences may be the typical mean length for single coherent thoughts in texts. We have also described several ways of visualizing the semantic structure of documents in space and time.

Mining the Temporal Structure of Thought from Text

In this paper, we present a signal-to-noise analysis of synchronous attractor networks of hysteretic threshold elements. The addition of hysteresis is known to enhance the capacity and convergence rate of attractor networks. The aim of this paper is partly to elaborate on results reported previously by other researchers, and to address the issue of whether there is an optimal value for hysteresis. Based on the simplified analysis reported here, we conclude that, for a given network loading (ratio of patterns stored to network size), there is an optimal value of hysteresis, but it changes as recovery proceeds to convergence. We hypothesize that a time-varying "hysteresis schedule" can be used to enhance the performance of attractor networks.

Ali A. Minai

Papers

Robust Deep Reinforcement Learning for Quadcopter Control.

Reading the Media’s Mind

Teaching Intelligent Systems at the University of Cincinnati

Mining the Temporal Structure of Thought from Text

Using hysteresis to improve performance in synchronous networks