Thursday, December 11, 2008

The art of forgetting: Bug or a feature?

- What would happen if we didn't forget a thing?

Our brain, through our senses, approximately consumes terabytes or more of information every second. Actually, the brain is a much faster processor. With about 50 billion brain cells firing about 200 times in a second, and activating about 100 other connected neurons in this firing (hypothetical), the brain can do a raw 1 million billion calculations per second!!! However, even a fine-grained decision making process may involve all these cycles in the brain as we are only able to take quick decisions (for example, image recognition) in about an order of a second time. (This is a very naive and hypothetical analysis based on lots of unsupported assumptions)

The point is that our brain has to reject and forget most of this stuff and retain the aggregated contained knowledge in its knowledge representation mechanism. This is not a bug but a very good feature! If we didn't forget a thing, we would get slower and slower over time and not only physically decay but also mentally decay and die as very 'slow' human beings (which is actually seen in some cases with people have some brain disorders)

There are several ways we forget. But again, the point here is that can we forget intentionally? Can we learn to forget? Take for example, a fish in an aquarium. A fish, as I can 'remember' has about ~5 seconds (or milliseconds?) of memory retention time (Time to live). Then it forgets everything and again starts exploring the aquarium as if she was in some new place. How bored would the fish become if she knew that small space enclosure very well and she had to spend all time there itself! 

The way these fishes permanently forget is actually different. If we can learn the way these fishes have the neuron structure and firings, we might gain some insights into the principles of forgetting. How would be just forget the things we remember to forget would be a very very interesting and challenging problem! "Did you remember to forget it?" :)

Desalination in Kidneys

I recently did a class project on desalination for the Biologically Inspired Design class. We explored several systems and came up with a really elegant design based on bio inspiration from intestines. I had dwelled on kidneys as another viable desalination option for the project. 

Kidneys are hugely complex structures that mix intelligent structure design with a 2 stage process for achieving their multiple purpose of filtering and maintaining the homeostatis balance. 

The great idea that kidneys use is the principle of investing energy on just actively selecting (reabsorbing) the good nutrients and elements back in the blood once it has been freely diffused to the urinary system (where all the work happens)

The idea is that of 'Selection' in favor to active 'Rejection'. Manmade systems (like dialyzers) generally come up with single process designs that either select or reject depending on the task at hand. 

This also helps in dividing the work into different components and not relying on the entire filtration and homeostatis process as a one step process where the kidneys somehow work and reject just the right amount of urinary fluid along with maintaining the ph balance. This rejection depends on how much of urine the body needs  out of its system. 

What is the difference between selection and rejection? If we look deeper in these processes, we may think of selection as the greedy process involving choosing the right amount and checking if the desired chosen quantity has been achieved. Whereas rejection in kidneys is a complementary and dumb process of diffusing out all the molecules that have smaller pores than membrane pores and meet the pressure conditions. A very complicated process would be involved if active selection had to be implemented just as a one step mechanism. 

Overall, I find this an amazing strategy for choice and selection of desired features. There can be several computational and algorithmic applications of this strategy. The vague computational analogy of this system is preprocessing the data and then extracting the desired features without search (as in a sorted index) (The other being just searching and extracting the desired elements to remove) 

Tuesday, December 09, 2008

Online social colonies as emergent systems

There was a recent article in Nature journal regarding how Google detected influenza epidemics using search engine query data. This was an amazing display of the power of technology to decipher emergent knowledge that could prevent the spread of an epidemic.  This was actually a very simple idea that can be abstracted to many new interesting unforeseen repercussion behaviors. 

Identifying social emergent behavior using collaborative online technology involves capturing implicit user communication that is meant for fixed tasks and outcomes. We have generally focussed on and studied the vocal communication aspect of humans. Man portends several other social communication behavior that are (mimetic, stridulatory, etc) equally sophisticated such as in other social species like bees, ants, dolphins, elephants and pigs!    

Some people are always interested to do things. Some want to do things because they like to work (like the workers of bee colony). Some people just follow other people. Some nag while the rest are either satisfied or scared and don't do anything. The communication pattern varies but has its own fixed role in a community.

The point is that there are implicit emergent behavior attached to all these activities as we are always in contact with other people and communicate with each other. Online technology is pushing this further in leaps because we now have the power to communicate with a very large number of people at once!  

Interdependence is a higher quality than independence. Fighting social problems as a group using technology (any problem that adheres to the 'Tragedy of the commons' principle and falls in the general public property category of problems - example - scarcity of water and other natural resources, etc) using implicit user behavior modeling and optimization and learning from our 'lesser intelligent' but other social species will result in cool outcomes!  

Thursday, December 04, 2008

Game-theoretic Modeling of Incentives between Healthcare Consumers and Providers

“The dollar isn't about expense, it's about selection and choice and commitment.” 

Healthcare information access is one of the top uses of Internet today. The internet now has matched physicians as the leading source for health related information. Emerging Web 2.0 sociable technologies are also enablers that promote collaboration amongst the regular people and surge healthcare information delivery. This way, the Internet is redefining the healthcare industry. Internet- based technology is headed towards delivering certain healthcare services more effectively and at lower costs. Newly emerging Internet healthcare stakeholders are poised to wrest control from established “brick-and-mortar” entities that have dominated healthcare for decades. This phenomenon is like the battle between the established Fourth Estate and new media technologies that are giving a hard time to the Press. Press is forced to re-think its survival strategies to sustain in the new Incentive based economic model that has dissipated control from a few fixed established houses to the general consumer and provider masses.


These unforeseen transformations need to be anticipated and modeled to benefit the stakeholders efficiently. The Internet has the potential to become the venue for the delivery of certain medical services, particularly those that are information-based. Virtual clinics are already starting to happen and involve online interaction between registered practitioners and patients. Currently, these services have a fixed cost per time financial model for consultation services and completely avoid the insurance companies from this interaction. In effect, this model is in its infancy and does not close the loop between healthcare service providers and delivery.  

Modeling this interaction effectively to analyze how the system performs is a very challenging task. This is more so important in this domain to avoid healthcare risks that may deteriorate the overall system by implementing not so thoughtful delivery channels. 

Game Theory studies strategic interactions between players. Given a set of players’ actions, and their preferences, game theory helps us analyze rational behavior between agents. The art of game theoretic analysis of a problem lies in the formulation of the problem as a matrix of players, strategies (actions) and the payoff(utilities) of the actions. Analyzing this matrix helps us discover best possible cooperating rational behavior among the players (Nash Equilibrium). Studying game-theoretic models (or other models that apply to human interaction) help us suggest ways in which an individual's behavior may be modified to improve one’s own welfare (for eg., applications in economics, decision theory and lately in computer science). By such analysis, the hope is to see the advantages and disadvantages of various strategies. 

More formally, a game is an interaction or a series of interactions between players, which assumes that 1) the players pursue well defined objectives (they are rational) and 2) they take into account their knowledge or expectations of other players behavior (they reason strategically). Players compete to maximize their payoffs with other players to reach an Equilibrium state.