Every indexed document

and using hierarchical architectures for reinforcement learning and sensory ab-

a new European project coordinated by Gianluca Baldassarre and conducted

spontaneously driven by intrinsic motivations. Kaplan

an initiative at the French research institute INRIA

the knowledge representations involved are neural network based

orchestrates actuator commands into complex movements

which makes it easy to map the design to massively parallel platforms

it can use feedback from DeSTIN’s action and critic networks to further

it will automatically form internal structures corresponding to the various

we can further simplify the expression such that

based on the past history of observations

and each of them comprises a certain ""spatiotemporal form"" recognized

each corresponding to a set of previously-observed

a statistical learning algorithm is used to predict subsequent states based on prior

which corresponds to a spatiotemporal region (nodes higher in the hierarchy corresponding

and attempt to construct belief states that capture regularities

and oftentimes pre-training. DeSTIN overcomes the limitations

this paradigm has its limitations; for instance

with the difference lying mainly in the DeSTIN control

general-purpose hierarchical control architecture. DeSTIN’s control hierarchy