LOCEN research (till May 2013)
Bio-constrained models
CLEVER-B4: modelling results of monkeys/children experiments and agency
Who: Francesco + Valerio + CNR-UCP, UCBM + USFD + AU + FIAS
What: Model data of experiments of children more closely, with focus on agency. Model: (a) IM based on hippocampus (habituator); (b) IM based on PFC (predictor). Scale the model to (a) make it work in denser spatial grid; (c) make it work with more meaningful actions: press, push left, push right, grap (?), etc.
CLEVER-B3: Goals
Who: Vincenzo, Francesco + USFD (Gurney)
What: Publish CLEVER-B3 with focus on different roles of dopamine: (a) DAt to trigger arm actions; (b) DAt to trigger goal permanent activation during learning; (c) DAp for repetition bias; (d) DAt to recall goals based Amg (CeA). But also more refined goal system: (a) in PFC goals have double activations: expected outcomes, intentions; (b) permanent activation of goals during learning; (c) goals are formed and recalled in anticipatory fashion by previous states.
Embodied attention + IM
Who: Valerio, Daniele, Simone Brun + FIAS + USFD Redgrave(?)
What: Models of gaze contingency paradigms with (a) realistic basal ganglia. Also with IM system based on: (b) luminance changes (SC); (c) Kohonen (Hip). Possibly with bio-constrained RL (yes?).
?Yes? Minimal model of Redgrave's theory
Who: Gianluca, Vieri (?), USFD (Redgrave)
What: Localist leaky-neuron model of IM, based on Redgrave view
RL-based models
IM + Goals
Who: Vieri, Marco
What: A model of goal selection and commitment for cumulative learning
TERL, IM, iCub
Who: Daniele, Valerio
What: (a) Stabilising TERL in iCub; (b) transfer to iCub; (c) add IM system; (d) show that the robot can learn in a cumulative fashion: task with moving objects.
Bruno multiple goal system on iCub
Who: Valerio, UMass (Barto, Castro da Silva)
What: ballistic task (e.g., throw of a ball?) on our iCub for testing Bruno's model
Students collaborations
Armando: Model of Hippocampus and novelty detection
Who: Armando Romani, Fancesco, John Lisman
What: Model Hip novelty detection system together with Lisman.
Emilio: Models of nucleus accumbens
Who: Emilio Cartoni, Puglisi/Cabib (?)
What: Bio-constrained model of nucleus accumbens: devaluation, specific PIT, aspecific PIT.
Francine: Models of intrinsic motivations in monkeys using virtual reality.
Who: Francine Dolins.
What: Start with pilot experiment based on current possiblity of the virtual reality program.
?Yes? Matthew: Models of intrinsic motivations to generate the passage of Watson
Who: Matthew Schlesinger.
What: Model of passage from attention to own body to attention to others; possible relations to autism.
?Yes? Edoardo: RL and IM (?)
Who: Edoardo Leopizzi (master student, from September, months ?)
What: RL model with intrinsic motivations, arm and eye controller, abstract arm actions.
PUBLICATIONS
- Paolo, Daniele: TERL
- Daniele: Berthier
- HA book: introduction
- HA book: hierarchical systems of LOCEN?
- Francesco: GDHL
- Francesco: Devaluation
- Vincenzo, Francesco: Stress coping
- Vincenzo, Francesco: CLEVR-B3, with focus on different dopamine functions
- Francesco: CLEVER-B4: focus?
- Bruno: model of throwing
- Daniele: BG and Cer with RL-EP and RL-DMP
- Vieri: model of goal formation and selection
- Daniele: TERL + IM
- Armando: model of hippocampus
- Francesco: Echo-state net started with Rob Low
- Francesco: Amg with Costanzi-Cestari
- Publication on IM-CLeVeR project
- Analysis of IM in good Journal
Major formal challenges for last year of IM-CLeVeR
- Win a new project (IDEAS? GBL?)
- Publish in outstanding journals
- Prepare last Periodic Project Report
- Organise IM-CLeVeR last Review Meeting with effective LOCEN iCub demo
- Organise Workshop on: "Intrinsic motivations and open-ended development in humans, other animals, and robots"
- Organise Frontiers in Cognitive Science number