Learning flexible sensori-motor mappings in a complex network

Given the complex structure of the brain, how can synaptic plasticity explain the learning and forgetting of associations when these are continuously changing? We address this question by studying different reinforcement learning rules in a multilayer network in order to reproduce monkey behavior in a visuomotor association task. Our model can only reproduce the learning performance of the monkey if the synaptic modifications depend on the pre- and postsynaptic activity, and if the intrinsic level of stochasticity is low. This favored learning rule is based on reward modulated Hebbian synaptic plasticity and shows the interesting feature that the learning performance does not substantially degrade when adding layers to the network, even for a complex problem.

Defining Goals and Learning Objectives for a Longitudinal Clerkship in the Complex Context of Health Care

In autumn 2007 the Swiss Medical School of Berne (Switzerland) implemented mandatory short-term clerkships in primary health care for all undergraduate medical students. Students studying for a Bachelor degree complete 8 half-days per year in the office of a general practitioner, while students studying for a Masters complete a three-week clerkship. Every student completes his clerkships in the same GP office during his four years of study. The purpose of this paper is to show how the goals and learning objectives were developed and evaluated. Method:A working group of general practitioners and faculty had the task of defining goals and learning objectives for a specific training program within the complex context of primary health care. The group based its work on various national and international publications. An evaluation of the program, a list of minimum requirements for the clerkships, an oral exam in the first year and an OSCE assignment in the third year assessed achievement of the learning objectives. Results: The findings present the goals and principal learning objectives for these clerkships, the results of the evaluation and the achievement of minimum requirements. Most of the defined learning objectives were taught and duly learned by students. Some learning objectives proved to be incompatible in the context of ambulatory primary care and had to be adjusted accordingly. Discussion: The learning objectives were evaluated and adapted to address students’ and teachers’ needs and the requirements of the medical school. The achievement of minimum requirements (and hence of the learning objectives) for clerkships has been mandatory since 2008. Further evaluations will show whether additional learning objectives need to be adopte

Reinforcement learning in dendritic structures

The discovery of binary dendritic events such as local NMDA spikes in dendritic subbranches led to the suggestion that dendritic trees could be computationally equivalent to a 2-layer network of point neurons, with a single output unit represented by the soma, and input units represented by the dendritic branches. Although this interpretation endows a neuron with a high computational power, it is functionally not clear why nature would have preferred the dendritic solution with a single but complex neuron, as opposed to the network solution with many but simple units. We show that the dendritic solution has a distinguished advantage over the network solution when considering different learning tasks. Its key property is that the dendritic branches receive an immediate feedback from the somatic output spike, while in the corresponding network architecture the feedback would require additional backpropagating connections to the input units. Assuming a reinforcement learning scenario we formally derive a learning rule for the synaptic contacts on the individual dendritic trees which depends on the presynaptic activity, the local NMDA spikes, the somatic action potential, and a delayed reinforcement signal. We test the model for two scenarios: the learning of binary classifications and of precise spike timings. We show that the immediate feedback represented by the backpropagating action potential supplies the individual dendritic branches with enough information to efficiently adapt their synapses and to speed up the learning process.

Reinforcement learning in dendritic structures

The discovery of binary dendritic events such as local NMDA spikes in dendritic subbranches led to the suggestion that dendritic trees could be computationally equivalent to a 2-layer network of point neurons, with a single output unit represented by the soma, and input units represented by the dendritic branches. Although this interpretation endows a neuron with a high computational power, it is functionally not clear why nature would have preferred the dendritic solution with a single but complex neuron, as opposed to the network solution with many but simple units. We show that the dendritic solution has a distinguished advantage over the network solution when considering different learning tasks. Its key property is that the dendritic branches receive an immediate feedback from the somatic output spike, while in the corresponding network architecture the feedback would require additional backpropagating connections to the input units. Assuming a reinforcement learning scenario we formally derive a learning rule for the synaptic contacts on the individual dendritic trees which depends on the presynaptic activity, the local NMDA spikes, the somatic action potential, and a delayed reinforcement signal. We test the model for two scenarios: the learning of binary classifications and of precise spike timings. We show that the immediate feedback represented by the backpropagating action potential supplies the individual dendritic branches with enough information to efficiently adapt their synapses and to speed up the learning process.

Gradient estimation in dendritic reinforcement learning

We study synaptic plasticity in a complex neuronal cell model where NMDA-spikes can arise in certain dendritic zones. In the context of reinforcement learning, two kinds of plasticity rules are derived, zone reinforcement (ZR) and cell reinforcement (CR), which both optimize the expected reward by stochastic gradient ascent. For ZR, the synaptic plasticity response to the external reward signal is modulated exclusively by quantities which are local to the NMDA-spike initiation zone in which the synapse is situated. CR, in addition, uses nonlocal feedback from the soma of the cell, provided by mechanisms such as the backpropagating action potential. Simulation results show that, compared to ZR, the use of nonlocal feedback in CR can drastically enhance learning performance. We suggest that the availability of nonlocal feedback for learning is a key advantage of complex neurons over networks of simple point neurons, which have previously been found to be largely equivalent with regard to computational capability.

Learning real-world stimuli in a neural network with spike-driven synaptic dynamics

We present a model of spike-driven synaptic plasticity inspired by experimental observations and motivated by the desire to build an electronic hardware device that can learn to classify complex stimuli in a semisupervised fashion. During training, patterns of activity are sequentially imposed on the input neurons, and an additional instructor signal drives the output neurons toward the desired activity. The network is made of integrate-and-fire neurons with constant leak and a floor. The synapses are bistable, and they are modified by the arrival of presynaptic spikes. The sign of the change is determined by both the depolarization and the state of a variable that integrates the postsynaptic action potentials. Following the training phase, the instructor signal is removed, and the output neurons are driven purely by the activity of the input neurons weighted by the plastic synapses. In the absence of stimulation, the synapses preserve their internal state indefinitely. Memories are also very robust to the disruptive action of spontaneous activity. A network of 2000 input neurons is shown to be able to classify correctly a large number (thousands) of highly overlapping patterns (300 classes of preprocessed Latex characters, 30 patterns per class, and a subset of the NIST characters data set) and to generalize with performances that are better than or comparable to those of artificial neural networks. Finally we show that the synaptic dynamics is compatible with many of the experimental observations on the induction of long-term modifications (spike-timing-dependent plasticity and its dependence on both the postsynaptic depolarization and the frequency of pre- and postsynaptic neurons).

Sustainable Wastewater Management: Is it Possible to Regulate Micropollution in the Future by Learning from the Past? A Policy Analysis

This paper applies a policy analysis approach to the question of how to effectively regulate micropollution in a sustainable manner. Micropollution is a complex policy problem characterized by a huge number and diversity of chemical substances, as well as various entry paths into the aquatic environment. It challenges traditional water quality management by calling for new technologies in wastewater treatment and behavioral changes in industry, agriculture and civil society. In light of such challenges, the question arises as to how to regulate such a complex phenomenon to ensure water quality is maintained in the future? What can we learn from past experiences in water quality regulation? To answer these questions, policy analysis strongly focuses on the design and choice of policy instruments and the mix of such measures. In this paper, we review instruments commonly used in past water quality regulation. We evaluate their ability to respond to the characteristics of a more recent water quality problem, i.e., micropollution, in a sustainable way. This way, we develop a new framework that integrates both the problem dimension (i.e., causes and effects of a problem) as well as the sustainability dimension (e.g., long-term, cross-sectoral and multi-level) to assess which policy instruments are best suited to regulate micropollution. We thus conclude that sustainability criteria help to identify an appropriate instrument mix of end-of-pipe and source-directed measures to reduce aquatic micropollution.

Firm-level strategic orientation and performance: a synthesis and extension of the knowledge base on entrepreneurial, market and learning orientation

Should a firm stay focused or should it rather adopt a broader strategic perspective? This dissertation summarizes and extends the existing knowledge base on entrepreneurial, market, and learning orientation. Building on multiple theoretical perspectives, empirical evidence from prior studies, as well as on survey and archival data collected in two economic contexts, performance effects from individual orientations, their dimensions and combinations are explored. Results reveal that the three strategic orientations are highly interrelated and that their relationship to firm performance is more complex than previously assumed.

Simulation Games - A Creative Tool for Capacity Building: Experience from the Learning for Sustainability Approach

Games that simulate complex realities to be dealt with in teams are an effective tool for fostering interactive learning processes. they link different levels of decision-making in the household, community and societal contexts. Negotiation and harmonisation of different perceptions and interests, be it within or between different households, form the basis of a common strategy for sustainable development.

Multimodal sequence learning

While sequence learning research models complex phenomena, previous studies have mostly focused on unimodal sequences. The goal of the current experiment is to put implicit sequence learning into a multimodal context: to test whether it can operate across different modalities. We used the Task Sequence Learning paradigm to test whether sequence learning varies across modalities, and whether participants are able to learn multimodal sequences. Our results show that implicit sequence learning is very similar regardless of the source modality. However, the presence of correlated task and response sequences was required for learning to take place. The experiment provides new evidence for implicit sequence learning of abstract conceptual representations. In general, the results suggest that correlated sequences are necessary for implicit sequence learning to occur. Moreover, they show that elements from different modalities can be automatically integrated into one unitary multimodal sequence.