top of page

It's a poor sort of memory that only works backwards

                                                    Lewis Carroll, 1871

   For diverse experiences (learnt gradually in time) to be exploited creatively in novel situations in a world (where neither everything is known nor can everything be experienced), it is necessary go beyond isolated studies related to learning different tasks and look at organization of memory, from the perspective of an integrated system that “perceives, acts, learns, reasons, makes mistakes, forgets” such interplay ongoing in an open ended fashion.

 

  Consider that, every touch, sight, taste, sound, movement triggers our memory system, enabling us to flexibly ‘connect the dots’ between our ‘past experiences’, the available ‘present’ and possible ‘future’. Simply, memory is not just ‘passive and reproductive’but prospective and future-oriented: our cognition an outcome of constructive manipulation of our memories, to benefit from experiences (of oneself, others). While the discovery of the Default Mode Network [1-2] now strongly indicates a common cortical network involved in recalling the past, simulating the future, goal directed planning and perspective taking [3], the underlying computational basis is blurred [4-5], with several open questions at the juncture of designing cognitive systems that can use experience flexibly in novel situations, understand the computational core prospective memory, in brain and machine.

 

iCub organization of memory

Prospective Memory Framework: Learning differnt things, at different times, by different means, and using experiences creatively in novel situations

In this context, our research on brainguided modelling of organization and use of memory in cumulativley learning systems [1,7-8] lies at the core of several pertinent topics:

Remembering: How past-learnt experiences are recalled based on present context (Objects, actions, sounds, people). The issue spans a large domain from designing Associative memories to medial temporal lobe modelling/understanding.

Abstraction/Consolidation: How are remembered past experiences combined with exploratory actions in the present to learn something new? How is Memory maintained, consolidated in a cumulatively learning system?

Prospection and Simulation: What computational mechanisms facilitate the recall of the past to simulate the future, take someone else’s perspective? What does the discovery of the Default mode network across species suggest about the underlying evolutionary constraints, shared problems, hence shared solutions?

Generativity/Creativity: How multiple past experiences are recombined to construct novel behaviors? 

Spatial/Non Spatial memory: Why should evolutionarily old networks engaged in spatial navigation like the hippocampus also be involved in memory? The urgent need to reconcile spatial navigation vs. memory function of the medial temporal lobe been emphasized recently by leaders in the field of memory [9]. The melting clocks of Salvador Dali, coincidentally called as Persistence of Memory, kind of depicts the close link between travlling in time and travelling in space! Being an avid follower of both Dali and his friend Rene Thom (whose catastrophe theory I use in the shape based Motor Skill framework)..I am also curious to know why Dali named a melting clock as Persistence of memory! back to shapes, Morphogenesis was also among Alan Turing's last works..(the published article in Royal Society is here!)

 

 

 

 

References, Further readings

  1. Mohan V, Sandini G, Morasso P. (2014). A neural framework for organization and flexible utilization of episodic memory in "cumulatively" learning baby humanoids, Neural Computation 26(12), 2692-2734, MIT Press.

  2. Sandini G, Mohan V, Sciutti A and Morasso P (2018) Social Cognition for Human-Robot Symbiosis—Challenges and Building Blocks. Front. Neurorobot. 12:34. doi: 10.3389/fnbot.2018.00034

  3. Buckner, R.L., Andrews-Hanna, J.R., and Schacter, D.L. (2008) The brain's default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci; 1124:1-38. [medline abstract] 

  4. Schacter, D.L., Addis, D.R., Hassabis, D., Martin, V.C., Spreng, R.N., Szpunar, K.K. (2012). The Future of Memory: Remembering, Imagining, and the Brain. Neuron, 76, 677–694.

  5. Mohan V,  Morasso P, Sandini G (2013). On the interplay between learning memory prospection and abstraction in a cumulatively developing system. IJCNN 2013, Dallas Texas, August 2013.

  6. Allen, T.A. & Fortin, N.J. (2013). The evolution of episodic memory. Proceedings of the National Academy of Sciences USA, 110 (Suppl 2), 10379-10386.

  7. Vernon,D, Beetz, M, Sandini, G. (2015) "Prospection in cognition: the case for joint episodic-procedural memory", Frontiers in AI and Robotics, Vol. 2, No. 19, 2015.

  8. Mohan V,  Morasso P, Sandini G,  Kasderidis S (2013) Inference through embodied simulation in cognitive robots. Cognitive Computation, Springer science publishers New York, DOI 10.1007/s12559-013-9205-4. 

  9. Mohan V., Bhat A.A., Sandini G., Morasso P. (2014) From Object-Action to Property-Action: Learning causally dominant properties through cumulative explorative interactions. Biologically Inspired Cognitive Architectures, Vol 10, 42-50.  http://dx.doi.org/10.1016/j.bica.2014.11.006.

  10. Bhat,A., Mohan, V., Morasso,P. Sandini,G. (2014). ‘’Connecting experiences’’: A biologically inspired episodic memory for developmental robots. 4TH IEEE ICDL-EpiRob, Genova, Italy.

  11. Eichenbaum H, Cohen NJ (2014). Can we reconcile the declarative memory and spatial navigation views on hippocampal function? Neuron. 2014 Aug 20;83(4):764-70.

Firstly, this video is around 7 mins, am sorry for that, but depicts some sequences of cumulative learning while the robot learns to construct the tallest possible tower with any random object (and combination) given to it in an open ended fashion. The details are in reference [4]. Several commonsense physical relations and affordances like large objects are better at the bottom, its rewarding to use mushrooms or pyramid like things at the top, color of things are not causally relevant, among others are learnt and organized in the memory..to genreate several novel creative behaviours...

bottom of page