Episodic memory, a term coined by Endel Tulving, is basically the memory system which provides the ability to recollect past personal experiences. It is that part of the memory which allows you to integrate the ‘what’, ‘when’ and ‘where’ features of personal experiences which can be recollected later. The episodic memory system is different from the semantic memory system although they collaborated closely for various activities.
The idea of building robots with an episodic memory system is an interesting area of research and we'll look at two recent papers on the subject and some of the problems and challenges of building an episodic memory system. But first, let’s look at some of the steps involved in building any memory system.
The stages involved in building an episodic memory system can be broken down as follows:-
(1) Encoding- This phase will deal with how the incident is captured. There are different ways of encoding this captured incident in the internal memory. A good encoding scheme could allow for relational clues among the different incidents captured
(2) Storage- The storage phase is concerned with how the captured memories are maintained. In this phase, there is an interesting problem to address, the question of which captured memories to delete when there the storage is full. Humans also forget things as time elapsed and so the this phase of design has to allow some memories to be ‘forgotten’. However, an exponential decay of memories would be a bad design as there may be some memories which needs to be preserved and are crucial for many other things. As we’ll see later, most of the existing solutions to this problem has not been satisfactory.
(3) Retrieval and use. The retrieval phase is concerned with retrieval of stored memory and the use phase is concerned with what or how the retrieved memory is going to be used.
The first paper - Deutsch,T., Lang,R., Gruber,A.,Velik,R. (2008) Episodic memory for autonomous agents.
This paper discussed most of the existing models on the subject and highlighted the major problems associated with each model. The paper also mentioned ISAC, a model with an interesting view of the role of emotion in remembering and forgetting memories.
Perhaps, the most interesting part is the speculation on the possibility of transferring the memories between semantic and episodic memory systems, to facilitate learning in robots.
The second paper is -
Kuppuswamy,N,S., Cho,S,E., Kim,J,H. (2006) A Cognitive Control Architecture for an Artificial Creature using Episodic Memory.
This paper described the cognition system of a virtual creature called Ritty. The virtual creature is a virtual model of a dog so there are many behaviours which are similar to a real dog. Ritty is placed in a virtual environment and its task is to explore the environment. The simulation showed that Ritty was able to optimize certain tasks by using the episodic memory system.
One of the main problems, which all of the models have is the problem of ‘forgetting’ as mentioned earlier. The ISAC humanoid has an emotion triggered mechanism which is an interesting solution, but this still mechanism still rely on an exponential decay mechanism indirectly. One can argue that the decay mechanism is sufficient for simple robots that perform simple tasks but as the tasks to be performed become more complicated, a more sophisticated mechanism is required.
The idea of building robots with an episodic memory system is an interesting area of research and we'll look at two recent papers on the subject and some of the problems and challenges of building an episodic memory system. But first, let’s look at some of the steps involved in building any memory system.
The stages involved in building an episodic memory system can be broken down as follows:-
(1) Encoding- This phase will deal with how the incident is captured. There are different ways of encoding this captured incident in the internal memory. A good encoding scheme could allow for relational clues among the different incidents captured
(2) Storage- The storage phase is concerned with how the captured memories are maintained. In this phase, there is an interesting problem to address, the question of which captured memories to delete when there the storage is full. Humans also forget things as time elapsed and so the this phase of design has to allow some memories to be ‘forgotten’. However, an exponential decay of memories would be a bad design as there may be some memories which needs to be preserved and are crucial for many other things. As we’ll see later, most of the existing solutions to this problem has not been satisfactory.
(3) Retrieval and use. The retrieval phase is concerned with retrieval of stored memory and the use phase is concerned with what or how the retrieved memory is going to be used.
The first paper - Deutsch,T., Lang,R., Gruber,A.,Velik,R. (2008) Episodic memory for autonomous agents.
This paper discussed most of the existing models on the subject and highlighted the major problems associated with each model. The paper also mentioned ISAC, a model with an interesting view of the role of emotion in remembering and forgetting memories.
Perhaps, the most interesting part is the speculation on the possibility of transferring the memories between semantic and episodic memory systems, to facilitate learning in robots.
The second paper is -
Kuppuswamy,N,S., Cho,S,E., Kim,J,H. (2006) A Cognitive Control Architecture for an Artificial Creature using Episodic Memory.
This paper described the cognition system of a virtual creature called Ritty. The virtual creature is a virtual model of a dog so there are many behaviours which are similar to a real dog. Ritty is placed in a virtual environment and its task is to explore the environment. The simulation showed that Ritty was able to optimize certain tasks by using the episodic memory system.
One of the main problems, which all of the models have is the problem of ‘forgetting’ as mentioned earlier. The ISAC humanoid has an emotion triggered mechanism which is an interesting solution, but this still mechanism still rely on an exponential decay mechanism indirectly. One can argue that the decay mechanism is sufficient for simple robots that perform simple tasks but as the tasks to be performed become more complicated, a more sophisticated mechanism is required.
No comments:
Post a Comment