IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information

Towards a Deep Improviser: a prototype deep learning post-tonal free music generator

“…The present Deep Improviser, while yet shallow, when comprised of CNN and RNN probably imitates some of the sequential aspects of time series models. But we also plan in the longer run to integrate Deep Improviser with algorithmic approaches based on information theoretic and perceptual decision-making models [46,47].…”

Linking Neural and Symbolic Representation and Processing of Conceptual Structures

“…IDyOT (Information Dynamics of Thinking: Wiggins, 2012b ; Wiggins and Forth, 2015 ; Forth et al, 2016 ) implements Baars' Global Workspace Theory (GWT; Baars, 1988 ), affording a computational model of a hypothetical cognitive architecture. At the functional level 1 , a number of generators sample from a complex statistical model of sequences (explained below), performing Markovian prediction from context (Wiggins and Forth, 2015 ; Forth et al, 2016 ). Each generator indexes a string of symbols, forming a chunk , a final substring of the overall memory model, expressed as symbols, whose origin is explained below.…”

“…Using this, a consolidation phase allows membership of categories to be optimized, by local adjustment, in terms of the predictive accuracy of the overall model. Theoretically, the layering of models and its associated abstraction into categories can proceed arbitrary far up the constructed hierarchy (Wiggins, 2012b ; Wiggins and Forth, 2015 ). Forth et al ( 2016 ) provide an account of the representation of timing in IDyOT; these aspects, however, are beyond the scope of the current paper.…”

“…The symbolic representation in our integration is that used in Information Dynamics of Thinking (IDyOT). IDyOT derives (e.g., sentence-like) structures by learning statistical sequential regularities over a linguistic corpus (Wiggins, 2012b ; Wiggins and Forth, 2015 ; Forth et al, 2016 ). IDyOT is unusual as a machine learning formalism in that it is symbolic in nature, but it generates and gives explicit semantics to its own symbols, in a bottom-up learning process, which is optimized by a general, data-independent principle of information efficiency, conceptualized as predictive accuracy.…”

Guest editorial

“…Following this, the research community has undertaken many studies exploring diverse facets of the subject, with more recent works and developments including Boden's 'The Creative Mind' 2 to the latest in artificial intelligence (AI) approaches to computational creativity such as exemplified by David Cope 3 and Geraint Wiggins. 4 Creativity research can be classified according to a global viewpoint embodied in romanticism, cognitive, social-cultural and systems approaches (see Childs and Fountain 5 ). Much current research work on creativity focuses on one or more of the four Ps, product, person, process and press.…”