PLoS Computational Biology journal has published Ten Simple Rules for Getting Published. My favorite is number 7:

7. Start writing the paper the day you have the idea of what questions to pursue

Your papers should be concise, and impart as much information as possible in the least number of words. Maintain a good bibliographic database as you go, and read the papers in it.

I'm using CiteULike and have my own list with 110+ articles and books, most of which I have read. Still, I am struggling with the fact that I don't have a good way of capturing my thoughts. A getting-things-done-like system may help, but its focus is on managing tasks and things to do, rather than a bunch of poorly organized and (not yet) related thougths and ideas. Yes, I have a paper notebook, but it doesn't capture granularity of thought I'm after. I also leave many notes on papers I read and no good place to transfer them to. The system covered in How to make a complete map of every thought you think looks very promising:

You'll really have clear idea of what kinds of thought are going through your head. You'll really understand your ideas. And you'll also see connections that you were never consciously aware of. You'll see a structure and a pattern in your life. Your goals and psychology will become clearer to you. You'll be clearer too about what you don't understand. You'll be familiar with your mental terrain. Addictive clarity. Vast clarity. Exceptional clarity. Extraordinary clarity.

The book covers materials, general principles, intra-, extra-subject architecture, theory of notebooks (!) and has everything to get you started. You just need to have patience to do a lot of work to keep it going.

Here is one more piece of advice on how to write a paper:

• Frame your issues right
• Argue for your issues
• Have a goal
• Have a structure
  • a. Goal: We present approach X for doing Y ....
  • b. Context: Others have tried Y, but all have problem P...
  • c. Support: The following data supports this observation
  • d. Approach: To do Y without suffering P, we do ...
  • e. Comparison: In comparison to others, we manage to ...
  • f. Conclusions: We succeeded in doing Y, without suffering P
  • Division of content: Abstract (30% a, 20% b, 30% d, 20% e+f). 1. Introduction (a+b) 2.
    Data (c) 3. Approach (d) 4. Comparisons (e) 5. Conclusions (f)
• Use the right structure/template
  • A design template (e.g. a design of a theory, an algorithm, a methodology...)
  • A problem template (for each solution we first need a good problem description!)
  • An evaluation template (of a set of approaches, resources, ...)
  • A research template (e.g. a linguistic phenomenon)

By the way, are you working on the most important problems in your field?

Unsolved problems of neuroscience:

• Awareness: What is the neuronal basis of subjective experience, wakefulness, alertness, arousal and attention? What is its function?
• Perception: How does the brain transfer sensory information into coherent, private percepts? What are the rules by which perception is organized? What are the features/objects that constitute our perceptual experience of internal and external events? How are the senses integrated? Is face perception special (e.g. inert)? What is the relationship between subjective experience and the physical world?
• Learning and Memory: Where do our memories get stored and how are they retrieved again? How can learning be improved? What is the difference between explicit and implicit memories? How plastic is the mature brain?
• Development: How and why did the brain evolve (the way it did)? What are the molecular determinants of individual brain development?
• Sleep: What is the function of sleep? Why do we dream? What are the underlying brain mechanisms? What is its relation to anesthesia?
• Cognition and Decisions: How and where does the brain evaluate reward value and effort (cost) to modulate behavior? How does previous experience alter perception and behavior? What are the genetic and environmental contributions to brain function?
• Language: How is it implemented neurally? What is the basis of semantic meaning?
• Diseases: What are the neural bases (causes) of mental diseases like psychotic disorders (e.g. mania, schizophrenia), Parkinson's disease, Alzheimer's disease or addiction? Is it possible to recover loss of sensory or motor function?

"Ten Unsolved Questions of Neuroscience", an upcoming book by David Eagleman and Patricia Churchland, presents a similar list (I couldn't find much about this book on any book site):

• How is information coded in neural activity?
• How are memories stored and retrieved?
• How do brains balance plasticity against retention?
• What is the brain’s extensive baseline activity about?
• How should we neurally characterize emotions?
• What is intelligence?
• How is time managed in nervous systems?
• Why do brains sleep and dream?
• How do the highly specialized subsystems of the brain coordinate with one another?
• What is the neural substrate of consciousness?

I have my own list of questions to think about.

Artificial eyes to "monitor areas over 360°" (I wish I knew how area over 360° looks like) in real time, touch sensors that allow feel out the likeness of Abraham Lincoln on a penny, and alcohol-powered muscles that are 100 times stronger than natural muscles, able to do 100 times greater work per cycle and produce, at reduced strengths, larger contractions than natural muscles. What's next?

[via Robots.net and Cognitive Daily]

Microsoft just announced Microsoft Robotics Studio -- a Windows-based environment that includes robotics development platform and lightweight services oriented runtime. The Studio provides simulation runtime and relies on the web services application model to interact with available services (this is especially close to my heart as the two books that I co-authored for O'Reilly were about webservices).

The toolkit currently supports LEGO Mindstorms (both RCX and NXT) and Fischertechnik interfaces.

The discussion group that covers Robotics Studio already has several interesting threads. From one of the messages (slightly edited for presentability); unfortunately I don't see any way to link to a particular message, but you can search for several "Absolute Horror" messages:

Robotics Studio will take your focus OFF the real problem - how to integrate timing into your design. Time is THE most important thing when you design a robot. You can have a dumb cap and resistor running the whole show as in Beam robotics and get great results. The further you abstract your framework and the more you rely on the simulation the more damage you will do to the community. What we really need is a BUS not a framework. And a way to hook up that bus to a PC. We do not need a framework without a bus. Robotics Studio is trying to create a distributed environment where timing will be explicit. These concepts are only applicable to the simple machines most people mistakenly call robots -- these are NOT robots. These are toys. [...] Instead learn the following:

• Everything you can get your hands on regarding oscillations and timing related to oscillations. Forget about traditional NNs - they are made of Layers (just like Ogres - watch shrek 1 for a good explanation) If they weren't made of layers (backpropagation is only good for non-temporal patterns) they just might work... Statistics (corrleation especially). Embodyment. Compliance. Types of learning (Supervised, Unsupervised, Reinforced). Hardware such as Bus design, servos, H-bridges, sensors. Forget PID control - your bot has to learn it otherwise it will not integrate well with the rest of the system.
• One thing I want to tell about amature robotics - the only way you get a complex behaviour is by increasing the number of actuators/sensors. The only way you can do it is by making the devices more affordable. The only way you can do that is by making the devices simpler and standardizing them. Robotics Studio wants to add XML processing to devices which contain 25-128 bytes of ram such as pics... and waist my precious CPU cycles on parsing it. I got one word: forgettaboutit.

Cofing4Fun also has a page dedicated to programming for LEGO Mindstorms (although no articles that cover the new NXT version yet). All this should run on top of Visual Studio Express, which is free.

Questions to ponder after reading of COGNITION-AND-REALITY, THE-QUEST-FOR-CONSCIOUSNESS, and THE-FEELING-OF-WHAT-HAPPENS.

• How does it happen that different people notice different aspects of the same situation?
• Why are some potions of the retinal input treated as belonging to the same object, others as independent?
  • May it have something to do with results of co-occurence analysis?
• Why do we often seem to perceive the meanings of events rather than their detectable surface features?
• How are successive glances at the same scene "integrated"?
• Why is perceiving almost always accurate, given inadequacies of the retinal image?
• If percepts are constructed, why are they usually accurate?
• What kind of cognitive structure does perception require?
• What happens when we choose what to see and how do we learn to see better?
• How are illusions and errors possible if perception is simply the pickup of information that is already available and specific?
• How do we pickup unanticipated information?
• What happens when a new object enters the field of view for the first time?
• How is it decided whether to use an existing schema or to develop a new one (exploitation vs. exploration)?
• How can anything at all be seen in a brief flash if perception is a temporally extended activity?
• How is schema modified by new information? What elements of it are being modified? Anticipation? Motor program?
• Is development of the schema from the general to the specific, from undifferentiated to precise or in the opposite direction?
• Why do introspective reports suggest that the meaning is available first, and the stimulus details only later or not at all?
  • This is probably because introspection starts from the highest level schema of all active schemata
• How does schema come to exist?
  • Some may be innate, but for those that are not, what's the mechanism? Co-occurrence?
  • The perceptual cycle must occur before it can develop
• Can we perceive something not having an appropriate schema?
  • Maybe only at the lowest possible level, which will allow for bootstraping based on some mechanism, like co-occurrence.
• Which schema develops earlier: the more complex "high-level" one, or the more simple "low-level" schema? (for example, shape vs. smile)
• How do schemata get co-activated? When they are related? Include one another? Not related?
• Does a schema for a concept (for example, "number") get easier and faster activated than a schema for an object that is part of that category (for example, number "five")? How do they influence each other?
• Is it possible to deploy more than one schema at a time?
  • Yes, multiple schemata can be active, even though they have different level of activation (and influence of other schemata); based on that level of activation they compete with other schemata for resources (gaze, effectors, and so on)
• What does presence of expected/unexpected information in the environment does to the schema? In terms of activation? In terms of its modification?
• Why is it difficult to pick up information from two different messages/streams at the same time?
• Is it possible to attend to two things at once?
• What are the limits of automatic mental activity?
• Why/how does perception depend on a skill? What skill is that?
  • Motor program that supports co-activation of other programs...
• Are schemata being influenced (activated/modified) by unattended information?
• Is there a single mechanism responsible for our cognitive limitations? What are the limitations? (THE-MAGICAL-NUMBER-SEVEN)
• What types of conflicts may arise from activation of two schemata? How do those conflicts get resolved/arbitrated?
• Is there some impediment to the parallel development of independent, but similar schemata? Does existing schema "canalize" incoming information and experience effectively preventing the second (similar) schema to be formed?
• Is consciousness an aspect of activity or an independently definable mechanism?
• Is it possible to think without being conscious?
• How does the subject know whether the present content of his consciousness originated with an external stimulus?
• Does imagery appear when pickup of information is delayed or interrupted?
• If images are anticipations rather than pictures, what's going on when we describe them?
• How can we imagine things we don't expect, for example, things we know cannot happen?
  • Anticipations concern things that only might come to pass rather than those things whose existence is already established
• How does detachment of images from the immediate context come about?
  • This detachment occurs inevitably in at least one situation with which we are all familiar: locomotion
  • Any delay between the anticipation and the pickup creates a state of unfulfilled perceptual readiness, and the inner aspect of that active shema is a mental image
• How are cognitive maps (and other types of schemata) acquired?
• What sorts of information do they incorporate at various stages in their development?
• How are they altered by experience?
• Under what conditions are they forgotten?
• Where many schemata exist, what distinguishes the right one?
• How schemata stored in memory? What represents long-term memory?
  • Repeated representation of the same material constitute a regularity to be detected
  • Use of schemata accounts for remembering
  • Forgetting occurs whenever the present inputis not specific enough to select a schema unequivocally
• What's the difference between perception and imagery?
  • Perception is a cyclic activity that includes an anticipatory phase; imagery is anticipation occuring alone
• What schemata are we born with? Schemata sensitive to expressions of emotion? Intentions?
• How words come to refer to objects?
  • Is it required to be "engaged in two perceptual cycles at once"? (COGNITION-AND-REALITY, p.164)
  • Are those words embedded in the schema or related to/associated with it?
• What is the principal function of grammatical structure of a sentence?
  • Is it to help the listener to develop proper anticipations that may span many seconds? (Ibid, p.167)
• How does one describe what one sees?
• Why do we feel that we know what we're going to say, but only in some general way?
• How does anticipations get detached from the stimulus information that generated them to become things we imagine?
• Who has more freedom in their actions: adult or infant?
• How do we perceive emotion of feeling in others?
  • Perceiver needs to have schemata to pick this information up
  • How much do these schemata depend on social experience?
• Why do we (sometimes) hear multiple voices arguing?
• How is anticipated image formed? What specifically is being anticipated (out of many possibilities)?
• How is movie-in-the-brain generated? How does the brain generate the sense of an owner and observer for that movie?
• What's the relationship between consciousness and emotion? (EMOTION-AND-CONSCIOUSNESS)
• Are there any non-conscious feelings? Are we conscious of all our emotions and feelings?
• What distinguishes emotion from feeling?
  • Emotions are external indication of feelings
  • Emotions are experienced as feelings
  • Emotions play signalling role; feelings facilitate learning and motivate behavior (on a larger scale) or produce a specific behavior (on a smaller scale) like freezing
    • "[Pleasure] is related to the clever anticipation of what can be done not to have a problem." "[Nature] seduces us into good behavior." (THE-FEELING-OF-WHAT-HAPPENS, p.78)
  • Expression of emotion drives feeling, which in turn supports expression of emotion; chicken and egg
• Can we control/supporess emotions/feelings?
• Why do we need to be conscious to have feelings or express emotions?
• What's the difference between sensing pain and knowning you have pain?
• What distinguishes zombies from normal (non-zombie) organisms? Automatism? Lack of intentions? Lack of continuity of purpose? Always-on learning (zombie learning)?
• What are the properties of consciousness?
  • William James (cf. THE-FEELING-OF-WHAT-HAPPENS, p.126): it's personal, selective, continuous, and pertains to objects rather than itself
• Why is there significant delay (up to 500ms; Libet's experiments) between registering of a sensation and conscious experience of it?
• How does always changing mind preserve its identity? What is is that provides mind with this core that is always the same? Is "self" a concept as any other concept and is built based on interactions with other individuals (culture and meme links)?

Modeling tools and simulators

• Genesis
• Neuron
• Conical (no longer under development)
• SpikeNet: a neural simulator for modeling large networks of integrate and fire neurons (C; source code available upon request?)
• spnet: a neural simulator for modeling large networks of neurons (C++)
  • SPNet++: modified version of spnet with MATLAB interface (C++)
• CSIM: A neural Circuit SIMulator (C++ code with MATLAB interface)
• SequenceNet: a neural network architecture for learning temporal information
• Amygdala: software for simulating spiking neural networks (C++)
• HHsim: Graphical Hodgkin-Huxley Simulator
• Neural Viewer (C++; no longer under development; no source code available)
• Biological Neural Networks (BNN) Toolbox (MATLAB)
• Alexandria
• Animal: a spiking neural network that learns to play chess (C++; no source code available)
• Metacat: a self-watching cognitive architecture for analogy-making and high-level perception
• Neural Networks at your Fingertips; not a model, but a collection of source code examples for various neural networks

What would I want from a framework or a simulator?

1. It needs to implement biologically plausible models
2. It needs to be fast enough to model tens of thousands on neurons with 1ms step (#1 and #2 usually don't come together)
3. It needs to provide an interface for developing simulations and monitoring their execution with real-time visualization (ideally as good as Neural Viewer does) and network activity diagrams
4. It needs to be easily configurable. All network parameters and their visualization should be compactly specified
5. It should support various sensory and motor systems
6. It should provide a comprehensive set of networks to play with

When I was a kid I liked playing blind chess with my dad and my brother. This is the type of chess when you make moves without looking at the board at all (you just tell your opponent what piece to move and where). I couldn't beat my father (to my defence I couldn't beat him very often in regular chess either), but I could play through end-game with 30 moves or more.

Now I want to compare this skill with my inability to remember a 10-digit phone number after hearing it one time or copy a 9-digit loan number between two computer screens from memory (I need to look at it at least twice). We all know about a span of immediate memory and its limit of seven plus or minus two items (THE-MAGICAL-NUMBER-SEVEN), which (partially) explains my difficulties with numbers. I just don't see how to reconcile it with my ability to play blind chess.

For those who never played blind chess, here is a brief summary of what you need to do. You need to 1) keep the current position in your head (initially 32 pieces on 64 squares), 2) be able to evaluate this position, and 3) consider possible moves and your opponent's responses (for any serious game this analysis needs to be done for several iterations/levels).

One possible explanation is that blind chess has more to do with mental imagery and less with working memory. While imagery is definitely involved here, I don't think this solves the problem. For starters, I don't imagine seeing the board as you'd see it in real life. The board that I see has no texture and no color; even squares are not colored black and white. I can "paint" it any color I want, but this will require some effort and "by default" it's completely feature-less. Much to my own suprise (this is the first time I examine this aspect of my own mental imagery), pieces too have no colors and even no shapes. I just "know" that this square has my pawn and that square has opponent's bishop. As with board colors, I can make them look anything I want, but this doesn't change the fact that by default they don't have any look. It's difficult to explain, but I just know conceptually that this is my piece on this square. Which makes me think that this has less to do with mental image and more with working memory that not only stores all those concepts and their relationships, but is also involved in analysis of the position and future moves.

Another possible explanation is that each of those tasks involves a different type of working memory; alternatively, there may be no separation, but working memory may have different task-specific capacity limits. For example, this article provides seven views on this capacity limit:

1. There are capacity limits but that they are in line with Miller's 7+2.
2. Short-term memory is limited by the amount of time that has elapsed rather than by the number of items that can be held simultaneously.
3. There is no special short-term memory faculty at all; all memory results obey the same rules of mutual interference, distinctiveness, etc.
4. There may be no capacity limits per se but only constraints such as scheduling conflicts in performance and strategies for dealing with them.
5. There are multiple, separate capacity limits for different types of material.
6. There are separate capacity limits for storage versus processing
7. Capacity limits exist, but they are completely task-specific, with no way to extract a general estimate.

Yet another explanation is that somehow I trained myself to store and process all this information. I don't buy this argument simply because I have been dealing with short sequences of digits for much longer than with blind chess, but without any significant progress.

I think it has something to do with the fact that "chunks" that I'm trying to remember in the first case (digits in a phone/loan number) are not related, while in the second case they are tightly related to each other. Even more importantly, what needs to be captured is a sequence of digits, rather than a group. This may be caused by mutual interference between elements in the sequence; as a result the sequence requires significant effort to maintain itself. If those number can be somehow related to each other (for example, they may be similar to a social security number) it may be much easier to remember by capturing "social security number with some modifications...". I think this also supports the idea of a hierarchical organization of (working) memory as while the chess play requires to maintain significantly larger number of elements, they all seem to be maintained "inside" that concept without much bear on working memory. Yes, you need to focus and pay attention when you play blind chess, but I'd argue that the effort required is probably less than when you need to remember a 9-10 digit sequence for 30 seconds.

Note that this effort doesn't seem to be proportional to the number of pieces to maintain in memory: it's easier to keep the position in memory in the beginning of the game when all 32 pieces are present rather than during end game where only few pieces may be left. This may be related to the fact that the initial position is more constrained (has fewer degrees of freedom) than any end game position, which makes it easier to track.

I'm watching Andrey who is trying to reach across a small table to grab his toy car. The car is too far and while the table is small he keeps trying even though he can't reach it and it would be very easy for him to go around the table and take the car. Still, he persists in using his (not working) method even after I tell him about the alternative.

This reminded me about what Douglas Hostadter called a difference between physical distance and problem distance (GODEL-ESCHER-BACH, p.612). When a dog wants to get his favorite bone that is just few feet away, but on another side of a fence with a gate not too far away he has two choices: 1) run up to the fence, stand next to it and bark; or 2) get to the open gate and double back to the bone. While the second options seems to be counterintuitive (as it increases the distance to the bone), it actually brings the dog closer to it.

What needs to "click" in dog's brain for this to happen? What's that new level of abstraction that allows to solve this problem? What does this solution in some abstract "problem space" have to do with the real material world, where this problem is being solved in a similar way?

Last time this happened six weeks ago. After brief one-time experience with a real fence Andrey knows how to fetch a toy he wants even though he may need initially to move away from his target. I continue being amazed how quickly he learns even from one-time experiences.

Chris Chatham posted a brief overview of the classic working memory model and offered a new diagram for that model. There are several interesting points, but I'd like to emphasize two of them: processing and memory are two sides of the same coin and gating functions are likely to be present at every intersection of arrows on the working memory diagram.

Having said that, I side with O'Reilly and his collaborators who proposed a completely different working memory model outlined in their article BANISHING-THE-HOMUNCULUS. The model is based on tripartite architecture, which is composed of the posterior cortex (PC) that performs majority of "automatic" sensory and motor processing, the hippocampus (HC) that is responsible for rapid learning that binds together arbitrary information, and the prefrontal cortex and the basal gandlia (PFC/BG) system that maintains internal contextual information (PFC), which can be dynamically updated by the BG.

In this model working memory is defined as an emergent property of the interactions between these three brain areas. These mechanisms not only support those basic memory functions that are generally associated with working memory, but also those controlled processing functions that are typically associated with a "central executive". The critical difference between this model and the model covered in Chris's post is that all information is viewed to be distributed in a relatively stable configuration throughout the cortext with working memory being represented by the controlled activation of those distributed representations. Authors' view is that working memory and executive function are two sides of the same coin, based on the fact that processing and memory functions are typically distributed within and performed by the same neural substrates.

As a side note, while a framework that separates working and long-term memory may look familiar and computantionally appealing, think about "simple" operation of copying a concept from long-term to working memory. What mechanism would support this copying that (to properly reproduce the concept) would need to include not only sensory and motor information associated with it, but also (potentially) large number of related concepts? How would "activation" of such a copy look like?

The authors also identify six key functional demands underlying working memory and describe how those are addressed in the proposed model:

• Rapid updating
• Robust maintenance
• Multiple, separate working memory representations
• Selective updating
• Top-down biasing of processing
• Learning what and when to gate

There is one more interesting point in the paper that I'll expand on later:

...it is likely that working memory may represent a kind of phylogenetic extension of the same kinds of mechanisms that underlie all forms of complex motor coordination and planning. (emphasis mine)

update 2006/03/20: More on working and long-term memory processes and assessment from Intelligence Testing.

Andrey is playing with his older brother, Daniil, who is 10 years old. Daniil is asking him to say words (in russian) -- "say, water; say, pipe; say, boat" -- and Andrey is repeating those words after his older brother. Now it's Andrey's turn: "say, (then after a 3 second delay, mimicking his brother) ...lamp; say, (after looking around) ...room; say, ...car" (much to our surprise as there wasn't any car around him). It looked like he was thinking about what to ask next and was doing search in his memory.

Overall, Andrey appears to be very proficient with words. He can easily say 4-syllable words, 4-5 word sentences, requests like "louder" (when listening to music in a car) and descriptions like "the smallest" (showing a small toy he is playing with).

Alysa's hand movements look more meaningful. She is not trying to reach for anything yet, but her movements look more purposeful maybe because they are longer, not as random as they used to be, and sometimes follow her eyes (probably just by accident). Her eye-hand coordination has also improved considerably. When she is awake she prefers positions where she can observe what’s happening around her.

First issue of Wired in 2006 published a list of The 50 Best Robots Ever. Unfortunately, my favorite robot, Johnny-5, was not on the list, even though the list includes many movie robots. I found it on The Ten Best Movie Robots list.

Out of those robots that were on the list, my favorite is Genghis, a six-legged walking robot created by Rodney Brooks and his team and covered with much detail in Brooks's book FLESH-AND-MACHINES. Its unique (for its time) feature was that it was based on the subsumption architecture (described in more detail in HOW-TO-BUILD-COMPLETE-CREATURES), which provides an incremental method for building robot control systems that link perception to action. This control system is implemented as a set of layers, with each layer progressively implementing more and more complex behavior; robot's macro behavior (following people) arises from many independent micro behaviors (moving legs, balance control, steering, and the like). Conflict resolution happens at the motor command level rather than at the sensor or perception level resulting in coherent, smooth, and, as Brooks calls it in his book, "lifelike" behavior. However, contrary to what the Wired article says, the robot doesn't learn; this behavior is "innate" and doesn't change with time.

Those of you who are interested in first hand experience, can start from one of the kits on the Top 10 Robot Christmas Gift Ideas compiled by the robots.net folks.

update 2006/03/20: Not really best robots, but still related to this topic. A brief overview of various military robots from Developing Intelligence

The functioning of an organism in a complex and unpredictable environment is supported by two complimentary processing systems, cognition and emotion. While the cognitive system is responsible for interpreting and making sense of the world, the emotion system is responsible for evaluating and judging events to assess their overall value with respect to the organism (ROBOT-EMOTION; p.2). Paul Ekman in Basic Emotions (HANDBOOK-OF-COGNITION-AND-EMOTION; ch.3) offers a slightly different perspective: "...the primary function of emotion is to mobilize the organism to deal quickly with important interpersonal encounters, prepared to do so by what types of activity have been adaptive in the past."

More specifically, functions of emotions in animals (including humans) can be summarized as follows (based on NEURAL-NETWORKS-AND-BRAIN-FUNCTION; pp.138-40):

1. Elicitation of automatic responses. Emotion may elicit an autonomic (for example, a change in a heart rate) or endocrine (for example, the release of adrenaline) response that prepares the body for action.
2. Flexibility of reinforcement. Emotional states allow a simple interface between sensory inputs and motor outputs, because only the valence of of the stimulus to which attention is being paid needs to be passed to the motor system, rather than a full representation of the sensory world. In addition to that, when a stimulus elicits an emotional state, we can flexibly choose any appropriate response, which is more flexible than simply learning a fixed behavioural response to a stimulus.
3. Motivation. For example, fear learned by stimulus-reinforcement association formation provides the motivation for actions performed to avoid noxious stimuli.
4. Communication. Animals can communicate their emotional state to others. What is the role of mirror neurons?
5. Storage of memories. Emotion may facilitate the storage of memories. This may be advantageous in that storing many details of the prevailing situation when a strong reinforcer is delivered may be useful in generating apropriate behavior in similar situations in the future. This may be facilitated in several ways: a) by bringing more attention to important situations the animal may be storing more information, b) by guiding attention toward what's important and away from distractions one may be storing more relevant information and c) by involving emotions stronger associations can be built.
6. Evaluation of memories. The current mood state can affect the cognitive evaluation of events or memories.
7. Memory recall: Emotion may trigger the recall of memories stored in neoroctical representations. Amygdala backprojections to the cortex could perform this for emotion in a way analogous to that in which the hippocampus could implement the retrieval of recent memories in the neocortex.
8. Better performance. The slowing of cognitive processes caused by negative emotions may enable more careful and deliberate scrutiny of self and circumstances, allowing the individual to gain a new perspective to help improve performance in the future. Negative effect allows us to think in a highly focused way while positive effect allows us to think more creatively and to make broader associations.

Paul Ekman in Basic Emotions identifies 15 basic emotions: amusement, anger, contempt, contentment, disgust, embarrassment, excitement, fear, guilt, pride in achievement, relief, sadness/distress, satisfaction, sensory pleasure, and shame. He also discusses enjoyment and surprise, but doesn't include those as basic emotions as they don't "subserve patterns of motor behavior which were adaptive for each of these emotions, preparing the organism for quite different actions." Also, interest "may be better regarded as a cognitive state rather than an emotion..." This list of emotions also doesn't include love, hate, grief, and jealousy as these are "emotional plots, more specific, more enduring than the basic emotions, specific contexts in which a number but not all of the basic emotions can be expected to occur." Ekman also defines moods -- which have different causes and last much longer, and are highly saturated with emotions -- and personaly traits, such as hostility.

Cynthia Breazel and Rodney Brooks in Robot Emotion: A Functional Perspective provide descriptions for a number of basic emotions that were implemented in the robot Kismet and gives details of that implementation (pp.17-20)

1. Anger. Anger serves to mobilize and sustain energy and vigorous activity at high levels. If it often elicited when progress toward a goal is hindered or blocked. This mobilizes an organism to try alternative strategies.
2. Disgust. Disgust is manifested as a distancing from some object, event, or situation, and can be characterized as a rejection of an unwanted stimulus.
3. Fear. The function of fear is to motivate avoidance of escape from a dangerous situation.
4. Joy. The emotion of joy is believed to heighten openness to experience. It often arises upon the success of achieving a goal or in the pleasure of mastery, exhibited even by very young children. The expression of joy operates as a universally recognizable signal of readiness for friendly interaction.
5. Sorrow. Experiencing of sadness is causing to slow the cognitive and motor systems, which enables one to reflect upon a disappointing performace to gain a new perspective that will help improve future performance. The expression of sorrow communicates to others that one is in trouble and increases likelyhood that the others will feel sympathy and lend assistance.
6. Interest. Interest motivates exploration, learning, and creativity. It mobilizes the creature for engagement and interaction. It serves as a mechanism of selective attention that keeps the creature focused on a particular object, person, or situation, and away from other distractions that impinge upon its senses.
7. Boredom. Boredom is an emotion that arises when the organism is not stimulated for a while. Hofstadter has an interesting note about boredom in GODEL-ESCHER-BACH (p.621): "you get bored with something not when you have exhausted its repertoire of behavior, but when you have mapped out the limits of the space that contains its behavior."

update 2006/03/02: Just noticed that Chris Chatham also posted some interesting (as always) information on this subject on his weblog earlier this month: Emotional Robotics and Simulating Emotion.

update 2006/03/21: This article by Edmund Rolls has an interesting diagram that put emotions on a scale from rage to relief and from terror to ecstasy.

Alysa maintains much better head control now and can turn her head while sleeping.

Alysa is well coordinating her eyes; this wasn't the case even two days ago.

I've been long puzzled by the fact that our brains can easily switch between working with classes and individual instances that represent those classes. Leveraging the idea that polychronous groups may represent symbols in the brain we can try to answer questions related to representation of classes and instances:

1. Are there symbols that represent classes, instances, or both?
2. Can a single symbol represent both, depending on how it's activated?
3. How do we define/measure "closiness" of two symbols? For example, is the sun closer to a tree than a submarine? Do we measure "closiness" in terms of activity/effort to undertake to go from one symbol to another one?

Jeffrey Elman in his papers FINDING-STRUCTURE-IN-TIME and AN-ALTERNATIVE-VIEW-OF-THE-MENTAL-LEXICON provides answers to these questions by using a model based on a simple recurrent network (SRN) where hidden unit patterns are fed back to themselves serving as the context for subsequent input patterns. The network was trained to predict the next word based on a corpus of sentences that were generated by a simple artificial grammar. The network was presented with words one by one and was tasked with predicting the next word. The network learned to predict words that were grammatically possible depending on the context.

Elman then looked at how the network categorized captured information. To analyze the similarity structure he averaged hidden unit activations for each word+context combination (after completing the learning phase) and then calculated Euclidian distance between generated vectors for each word. The network learned to partition network's "mental" space into major categories, which were further subdivided into smaller categories (like humans or food).

There are several interesting things about this category structure.

1. It appears to be hierarchical.
2. This structure is "soft" and implicit, with some categories being quite distinct and others having less distinct boundaries and sharing properties with other categoires.
3. The content of the categories and their structure is not known to the network.
4. Network's representations are highly context-dependent, which means that there could be separate representations for the same word that occurs in every different context. All concepts are expressed in a distributed manner as activation patters over a fixed number of nodes. A given node participates in representing multiple concepts. The activation of an individual node may be uninterpretable in isolation -- it is the activation pattern that is meaningful in its entirety. (FINDING-STRUCTURE-IN-TIME; p.21)

One can clearly see some parallels between this description and ideas expressed in the previous post on symbols and polychronous groups.

Capturing information about instances rather than classes provides even more benefits because "this notion of categories as emergent from the location in a high-dimensional state space, in which at any given moment different dimensions might be attended to and others ignored, suggests that different viewing perspectives on that space might yield new categories." (AN-ALTERNATIVE-VIEW-OF-THE-MENTAL-LEXICON; p.4) Also, once in place, this category structure supports generalization as every new word, identified as belonging to an existing category, inherits all the properties that are assigned to that category. Compare this with the prototype principle expressed in GODEL-ESCHER-BACH (p.352) as:

The most specific event can serve as a general example of a class of events.

You may notice that all those categories that "emerge" as the result of this hierarchical clustering are not explicitly represented in the model; an external observer attaches a label of "animates" or "inanimates" to groups of instances that represent those classes. How would a label like this, for example, "animates" be represented? I think it would not necessarily belong to the same space that is occupied by the instances in the class it represents; rather its relatioships with items in that class or category would be rule-based ("I know if something moves on it's own, it's part of the animates class").

Elman's paper doesn't specifically address the related case of category splitting: how do we acquire "labrador" or "cocker spaniel" category after acquiring the "dog" category? From that we can go to "Snoopy", which may be an instance of the "cocker spaniel" category, or even to "my dog Snoopy".

Overall, it seems like even a simple recurring network can produce quite complex results. It might be interesting to replace hidden and context units with polychronous groups and see whether it's possible to reproduce Elman's results. I see at least two challenges with using the group approach: one is proper assignment of input and output units (how do you read the state "group A is active") and the second is measurement of "closiness" of two groups to do clustering analysis.

So far I've been focusing on the modeling of individual neurons and their networks, but after reading more on throught processes in GODEL-ESCHER-BACH (pp.348-50), I thought it might be interesting to look at the level of neural complexes that represent symbols in the brain, rather than at the level of individual neurons and signals between them.

There are many questions that come to mind concerning symbols and their representations in the brain:

1. How many neurons need to be connected to form a neural complex?
2. Can a neuron belong to more than one complex?
3. To how many complexes can a single neuron belong?
4. How many neurons need to be activated for the symbol to become active?
5. Are there certain core neurons for each symbol that become active every time the symbol is activated?
6. How many neurons can two symbols share?
7. How similar are complexes that represent the same symbol in different brains?
8. What regions of the brain these complexes extend into?

Then Hofstadter offers a hypothesis about symbol representation in the brain (pp.356-7):

...overlapping and completely tangled symbols are probably the rule, so each neuron, far from being a member of a unique symbol, is probably a functioning part of hundreds of symbols.

...in order to distinguish one symbol's activation from that of another symbol, a process must be carried out which involves not only locating the neurons that are firing, but also identifying very precise details of the timing of the firing of those neurons. Thus perhaps several symbols can coexist in the same set of neurons by having different characteristic neural firing patterns.

These ideas come very close to ideas expressed by Izhikevich in his paper on polychronization, which is defined as reproducible time-locked but not synchronous firing patterns with millisecond precision. Neurons are spontaneously organized into polychronous groups by the spike-timing-dependent plasticity (STDP) changes that select conduction delays to allow the groups to form. As each neuron participates in many groups, firing with one group at one time and with another group at another time, the number of coexisting polychronous groups could be far greater than the number of neurons in the network and even greater than the number of synapses.

What is the significance of polychronous groups? We hypothesize that polychronous groups could represent memories and experience. (p.261)

...the system has potentially enormous memory capacity and will never run out of groups, which could explain how networks of mere 10¹¹ neurons (the size of the human neocortex) could have such a diversity of behavior. (p.270)

Let's entertain the idea that polychronous groups are, in fact, neural complexes described above that represent symbols in the brain. Now we can try to answer questions on the list above:

1. How many neurons need to be connected to form a neural complex? Probably as few as five. Izhikevich's paper shows five neurons with optimized delays that generate 14 groups.
2. Can a neuron belong to more than one complex? Yes.
3. To how many complexes can a single neuron belong? Potentially more than the number of synapses it has.
4. How many neurons need to be activated for the symbol to become active? From the paper (p.267): "When a group is activated, whether in response to a particular stimulation or spontaneously, it rarely activates entirely. Typically, neurons at the beginning of the group polychronize, that is, fire with the precise spike-timing pattern imposed by the group connectivity, but the precision fades away as activation propagates along the group. As a result, the connectivity in the tail of the group does not stabilize, so the group as a whole changes."
5. Are there certain core neurons for each symbol that become active every time the symbol is activated? According to this model, there are always some neurons that need to be activated for a group to become active.
6. How many neurons can two symbols share? I don't think there are any limits. Two groups can share all their neurons and have separate activation patterns.
7. How similar are complexes that represent the same symbol in different brains? Not similar at all.
8. What regions of the brain these complexes extend into? I can't give you any definitive answer, but probably not all. Can it be that we have two types of regions in the brain: some that are formed with specific neuronal circuits and others that have randomly connected networks of neurons that self-organize into groups based on their inputs?

After thinking about some of the answers I came up with more questions:

1. Can groups be replicated? It seems like all we need to do is to provide the same input and then the group(s) emerge after STDP selects proper delays. It may not be the same group, but it will represent the same concept in a different place in the brain. Or will it? Is a symbol defined by its neuronal representation or by its connections with other symbols?
2. How do we think of something we never saw before? How do we connect those symbols to form a new concept? Do they need to form one group? Do they need to be active at the same time? How do we imagine three dogs in a teacup? Do we imagine three dogs somewhere else and then make it look like a teacup when we focus our attention on it? More on this from GODEL-ESCHER-BACH (p.362): "As we imagine a hypothetical event, we bring certain symbols into active states -- and depending on how well they interact (which is presumably reflected in our comfort in countinuing the train of thought), we say the event 'could' or 'could not' happen."
3. How do we keep symbols active? How do we morph them into something they are not? Do we create a new symbol based on the existing one every time we do this or do we temporarily change the existing symbol?

It's also interesting to note that, while any particular neuron can be a part of many groups, it can't participate in all those groups at the same time; at any given moment it can only be a member of at most one (?) group.

Try to guess what the letters mean.

Intelligence test, part 1:

• 24 H in a D
• 26 L of the A
• 7 D of the W
• ... and 30 more

Intelligence test, part 2:

• 5 to 7 C of the W
• 1 O G E 4 Y
• ... and 22 more

Definitely fun, but doesn't look like a substitute for the Turing Test.