Local Self-Adaptation Mechanisms for Large-Scale Neural System Building

Ortíz, Matadamas; Gepperth, Alexander

doi:10.1007/978-90-481-9695-1_83

Cited by 2 publications

(2 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Such learned kernels would implement a true data model and could thus realize far more interesting and useful operations than shown here. And lastly, the ad hoc way of choosing input transfer functions for boosting input strength to the region where they will actually create activity in subsequent layers needs evidently to be replaced by an automatic adaptation process of a slow, homeostatic nature as, e.g., outlined in [28] and implemented in [27].…”

Section: B Shortcomings Of This Studymentioning

confidence: 99%

“…In addition to the original model, we include a point-wise applied input transfer function f I [S] which is a tool to bring the sum of inputs into a value range where it can excite the field effectively. f I will be replaced by a homeostatic self-adaptation process in the future, such as proposed in [27]. Here, the goal is to amplify weak inputs somewhat but to limit input values to [0, 1].…”

Section: B Single-layer Model Equationsmentioning

confidence: 99%

See 1 more Smart Citation

Latency-based probabilistic information processing in a learning feedback hierarchy

Gepperth

2014

2014 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

In this article, we study a three-layer neural hierarchy composed of bi-directionally connected recurrent layers which is trained to perform a synthetic object recognition task. The main feature of this network is its ability to represent, transmit and fuse probabilistic information, and thus to take near-optimal decisions when inputs are contradictory, noisy or missing. This is achieved by a neural space-latency code which is a natural consequence of the simple recurrent dynamics in each layer. Furthermore, the network possesses a feedback mechanism that is compatible with the space-latency code by making use of the attractor properties of neural layers. We show that this feedback mechanism can resolve/correct ambiguities at lower levels. As the fusion of feedback information in each layer is achieved in a probabilistically coherent fashion, feedback only has an effect if low-level inputs are ambiguous.

show abstract

Section: B Shortcomings Of This Studymentioning

confidence: 99%

Section: B Single-layer Model Equationsmentioning

confidence: 99%