The intervention of technology has made immense developments in both functional domains
of humanity- communication and Computational Intelligence. Communication begins with cave paintings and now we are enjoying smart phone and social networks. In the field of computation, we have computers with processor- Intel 13 th generation Core i9. With 6.0 GHz processor speed, Supercomputers. And now going to have quantum computers.
These technological advancements and breakthroughs have happened by virtue of the
intelligence of the human being. The initial goal to bring in these advancements was ease
and comfort. And of late, focus has been on enhancing the human productivity. The people
further began to explore if machines precisely speaking computing and communication
machines can solve the problems which are reserved for human being.
We know that the computer is made to work through programming. There is an instruction
set to tell the computer what to do and how to do? In first instance, it does not seem to be
having cognitive abilities. Along with time, a debate began on “ Can computer think?
In 1950, Alan M Turing, an English mathematician proposed ‘Turing Test.’ The purpose was
to examine if “ computer can think? He believed, “ There is nothing which a person can do
and a computer can not do.
We all know that ‘Thinking’ is a cognition ability. And with this begins a mission to build
computing machines which can act and think like human being. Such a machine is known as
intelligent machine. This is to note that intelligence is the capacity of thinking and reasoning.
In 1956, a workshop on Dartmouth Summer Research Project on artificial intelligence(AI)
was organized. John McCarthy -an American computer scientist was the key person to
organize this workshop- the founding event of AI. John McCarthy is recognized as the Father
of AI.
In simple language, AI is the emulation of human intelligence on machines. So that the
machine can act and think like human being. Another definition, “AI is the study of how to
make computers do the things at which people are doing better”.
Emulation of human intelligence needs understanding the human brain. So, AI necessarily
involves modeling of the biological process of the brain- precisely speaking – modeling of the
biological neural system. We shall all agree that the most important attribute of intelligence is Learning. In fact, Learning is what an entire intelligent system does.But how exactly learning happens in human brain, is indeed a very complex biological task. Not yet understood.
How to empower the machines with the skills of learning?
It is a big challenge. There is one way out to understand this. Probably people memorize situation-action- pair in the brain. In the learning process: process of encoding the situation-action pairs on the
memory.
In the event of a situation, memory should recall the correct action.
One way out to understand the process of learning in machines is through training the
machines with known situation-action pairs.Computational intelligence is essentially a subset of AI. AI is based on hard computing techniques. CI is based on soft computing methods. Hard computing is driven by analytical models. Soft computing is heuristic in nature. Example, Fuzzy systems, evolutionary
computing, and neural networks.
These are capable of handling imprecision and uncertainty.
The very important point here to note that human being has the ability to reason and learn in
the environment of imprecision and uncertainty. CI approach is closer to natural intelligence.
Prof. James Bezdek defines the computational intelligence:
A system is computationally intelligent when it deals with numerical data, has pattern recognition components, does not use knowledge in AI sense, and additionally it exhibits
(1) Computational adaptability
(2) Fault tolerance
(3) Speed and error rates approaching the human performance
Looking the requirements and characteristics of CI, we may outline four paradigms of CI:
- Neural networks
- Evolutionary computing
- Swarm intelligence
- Fuzzy system
Interestingly, all the paradigms have their origins in biological system. Artificial Neural
network is a model of the biological neuron. 10-500 billion neurons in human cortex, 60
trillion synapses, 1000 main models each with 500 neuron networks, modeling of ANN of
such a large configuration is too much challenging.
2020 onwards deep learning algorithms based on multilayer neural networks are being
extensively investigated for speech recognition, natural language understanding and computer
vision. Important NN structure include RNN, LSTM, and CNN. Training relies on the
backpropagation by computing gradient loss function and then update the weights. Gradient
vanishing problem limits the number of layers and training speed.
The most recent development in NN is “Incremental Learning.” Ability to identify the class example
belonging to new class. It reduces the error of misrecognition of known classes. Less
resources are used and classification accuracy is enhanced to good level. It is an interesting
research problem how to design ensemble layer to identify novel classes. Modify the
SoftMax layer. Such open set classification is finding applications in NLP.
Evolutional Computing involves natural evolution; the concept is survival of the fittest, the
weak must die. Natural evolution ; success is achieved through reproduction. Genetic
algorithm is an excellent class of EC. Swarm intelligence is based on the social behavior of organism living in swarms or colonies. Study of colonies; efficient optimization and clustering algorithms; choreography of bird’s flocks led to the design of particle swarm optimization; foraging behavior of the ant’s colony optimization.
Particle Swarm Intelligence, a global optimization approach is based on the
social behaviors of birds’ flock. The fuzzy system is the study of how organisms interact with their environment. It is like human reasoning; always not so exact as true or false, 1 or 0.
The fuzzy sets and logic allow approximate reasoning. It enables modeling of the common
sense. CI is the most rapidly evolving area with varied applications in sciences and society.
Though CI has numerous applications.
These include robotics, image processing, NLP, autonomous navigation, medical diagnosis, Fault tolerance, Decision support systems- stock
market predictions, risk assessment in finance, recommendation for e-commerce.
Computational intelligence is a powerful and promising tool to offer quick solution to our
complex problems.
The need of the hour is to develop the CI models trained on the data collected from indigenous implementations. It is certain CI and AI will dominate the next 10
years trajectory of social movement. Let us become a contributory agent of this
transformation.
