Studies of Goal Directed Movements by Emanuel V. Todorov

The main idea [of the discussion on a Bayesian model of sensory-motor processing] was that the sensory-motor system uses a casual model to predict the outcomes of hypothetical actions, and a generative model to predict sensory stimuli arising from hypothetical states of the world. These two models are then "inverted" via some powerful computation. [The Kalman filtering algorithm] does exactly that: the first line is the casual model describing the hand dynamics, and the second line is the generative model describing how sensory inputs arise. (p. 29)

Motor templates and learning

Memory decay and target resampling

Imagine that for some reason the controller is using inaccurate estimates of the noise terms in the system. In particular it is too confident in the sensory input (G is smaller than the actual amount of sensory noise) and too skeptical about stability of the world (D is larger than the actual amount of additive system noise). Such a controller will use a Kalman gain K larger than the optimal value i.e. it will overcorrect based on sensory inputs; in particular, in the absence of inputs the gain will not become exactly 0 and thus the state estimate (memory) will gradually degrade. Furthermore, the estimated variance of the state will be larger than its true value, thus from the point of view of the controller it will be advantageous to sample the sensory input as often as possible, and in particular look at the targets even if they have been presented before the movement. (pp. 38-9)

Sensory adaptations

Using a default control law L that corrects for systematic perturbations detected on previous trials can have effects quite similar to sensory adaptations. (p. 39)

It is possible that the primary objective of the adaptation process is restoring performance of the task (i.e. acquire the target withing the time limit) rather than restoring the shape of the baseline trajectory -- the latter being an epiphenomenon. (p. 93)

Emotion and Consciousness: End of a continuum
Yuri I. Alexandrov, Mikko E. Sams
Cognitive Brain Research 25 (2005) 387-405

Banishing the Homunculus: Making Working Memory Work
Hazy, T.E., Frank, M.J. & O'Reilly, R.C. (2006)
Neuroscience.

An important paper that outlines six key functional demands underlying working memory and presents a working memory model based on representations in the prefrontal cortex which are dynamically gated by the basal ganglia (PBWM). Working memory is described as controlled activation of stable configurations that exist throughout the cortex (rather than a separate module with information being moved between long-term memory and working memory bufers).

A unifying view of the basis of social cognition (copy)
Vittorio Gallese, Christian Keysers and Giacomo Rizzolatti

In this article we provide a unifying neural hypothesis on how individuals understand the actions and emotions of others. Our main claim is that the fundamental mechanism at the basis of the experiential understanding of others’ actions is the activation of the mirror neuron system. A similar mechanism, but involving the activation of viscero-motor centers, underlies the experiential understanding of the emotions of others.

These studies suggest that the activity of mirror neurons correlates with action understanding. The sensory features of the perceived actions (partially seen or just heard) are fundamental to the activation of mirror neurons only inasmuch as they trigger the motor representation of the same actions within the observer/listener brain.

...these data show that the human motor system codes both the goal of an observed action and the way in which the observed action is performed.

...these data indicate that when we see someone performing an action, besides the activation of various visual areas, there is a concurrent activation of part of the same motor circuits that are recruited when we ourselves perform that action.

...a similar mechanism is also involved in our capacity to understand and experience the emotional states of others.

An Alternative View of the Mental Lexicon
Jeffrey L. Elman
(2004)

Finding Structure in Time
Jeffrey L. Elman
Cognitive Science, 14, 179-211 (1990).

Unsupervised learning of natural languages (copy)
Zach Solan, David Horn, Eytan Ruppin, and Shimon Edelman (2005)
PNAS | August 16, 2005 | vol. 102 | no. 33 | 11629-11634

We address the problem, fundamental to linguistics, bioinformatics, and certain other disciplines, of using corpora of raw symbolic sequential data to infer underlying rules that govern their production. Given a corpus of strings (such as text, transcribed speech, chromosome or protein sequence data, sheet music, etc.), our unsupervised algorithm recursively distills from it hierarchically structured patterns.

This unsupervised algorithm is capable of learning complex syntax, generating grammatical novel sentences, and proving useful in other fields that call for structure discovery from raw data, such as bioinformatics.

A synapse which can switch from inhibitory to excitatory and back
Erik Fransén (2005)
Neurocomputing: 65-66, 39-45

In this work we are investigating the functional consequences for a synapse if it had both co-release and conditioning depression. If initially the GABA component is larger than the glutamate component, the synapse has an inhibitory net effect. However, if the postsynaptic cell is conditioned, the GABA component will be suppressed yielding an excitatory synapse.

Simple Model of Spiking Neurons
Eugene M. Izhikevich
IEEE Transactions on Neural Networks (2003) 14:1569-1572

A model is presented that reproduces spiking and bursting behavior of known types of cortical neurons. The model combines the biologically plausibility of Hodgkin–Huxley-type dynamics and the computational efficiency of integrate-and-fire neurons. Using this model, one can simulate tens of thousands of spiking cortical neurons in real time (1 ms resolution) using a desktop PC.

A Quantitative Map of the Circuit of Cat Primary Visual Cortex
Tom Binzegger, Rodney J. Douglas, and Kevan A. C. Martin

The map of the whole cortical circuit shows that there are very few “strong” but many “weak” excitatory
projections, each of which may involve only a few percentage of the total complement of excitatory synapses of a single neuron.

Conceptual precursors to language
Susan J. Hespos, Elizabeth S. Spelke

[Findings of this paper] provide evidence that infants use the tight–loose distinction in predicting object motion: they infer that motion of a contained object will cause a conjoint, rigid motion of the container if, and only if, the object and container fit tightly. Because non-human primates display similar capacities14, infants’ action categories seem to be linked to a language-independent system for representing objects, rather than to any representation specific to the language faculty. When language evolved as a system for linking sounds and concepts, it probably built upon a repertoire of pre-existing conceptual capacities.

Language learning ... seems to develop by linking linguistic forms to universal, pre-existing representations of sound and meaning.

Segregation of the Brain into Gray and White Matter: A Design Minimizing Conduction Delays
Quan Wen, Dmitri B. Chklovskii
(2005) PLoS Comput Biol 1(7): e78.

A ubiquitous feature of the vertebrate anatomy is the segregation of the brain into white and gray matter. Assuming that evolution maximized brain functionality, what is the reason for such segregation? To answer this question, we posit that brain functionality requires high interconnectivity and short conduction delays. Based on this assumption we searched for the optimal brain architecture by comparing different candidate designs. We found that the optimal design depends on the number of neurons, interneuronal connectivity, and axon diameter. In particular, the requirement to connect neurons with many fast axons drives the segregation of the brain into white and gray matter. These results provide a possible explanation for the structure of various regions of the vertebrate brain, such as the mammalian neocortex and neostriatum, the avian telencephalon, and the spinal cord.

Polychronization: Computation With Spikes.
Eugene M. Izhikevich
Neural Computation (2006) 18:245-282

The major result of this article is that spiking networks with delays have more groups than neurons. Thus, the system has potentially enormous memory capacity and will never run out of groups, which could explain how networks of mere 10¹¹ neurons (the size of the human neocortex) could have such a diversity of behavior. Of course, we need to learn how to use this extraordinary property in models. (p.270)

Which Model to Use for Cortical Spiking Neurons?
Eugene M. Izhikevich
IEEE Transactions on Neural Networks (2004) 15:1063-1070

If the goal is to study how the neuronal behavior depends on measurable physiological parameters, such as the maximal conductances, steady–state (in)activation functions and time constants, then the Hodgkin–Huxley type model is the best. Of course, you could simulate only tens of coupled spiking neurons in real time.

In contrast, if you want to simulate thousands of spiking neurons in real time with 1 ms resolution, then there are plenty of models to choose from. The most efficient is the I&F model. However, the model cannot exhibit even the most fundamental properties of cortical spiking neurons, and for this reason it should be avoided by all means.

If the goal is to understand the fine temporal structure of cortical spike trains, and to use spike-timing as an additional variable to understand how the mammalian neocortex processes information, a spiking model that can exhibit all or most of the 20 neuro-computational properties of biological neurons is required. The model recently proposed by Izhikevich was developed exactly for these purposes. It is the simplest possible model that can exhibit all the firing patterns in Fig. 1.

A model of STDP based on spatially and temporally local information: Derivation and combination with gated decay (copy)
Anatoli Gorchetchnikov, Massimiliano Versace, and Michael E. Hasselmo
Neural Networks, Volume 18, Issues 5-6, July-August 2005, Pages 458-466

This paper presents a simplified spike-timing-dependent plasticity (STDP) learning rule as well as five types of gating derived from conventionally used types of gated decay in learning rules for continuous firing rate neural networks.

Relating STDP to BSM
Eugene M. Izhikevich and Niraj S. Desai
Neural Computation (2003) 15:1511–1523

This paper demonstrates that a standard LTP/LTD learning rule called the BCM (Bienenstock-Cooper-Munro) follows directly from spike-timing-dependent plasticity (STDP) when pre- and postsynaptic neurons fire uncorrelated or weakly correlated Poisson spike trains, and only nearest-neighbor spike interactions are taken into account, which makes this approach very computationally efficient.