Machines Learn Like Humans

Scientists create machines that mimic
the activity of human brain, reports Biplab Das 

Be it an exposure to odour, colour or a particular shape, human brain is tailored to learn from such encounters. Simulating this unique ability of the human brain, scientists have created a learning technique called artificial neural network, which works like human neurons. "In the human brain, a typical neuron collects signals from other neurons through a host of fine structures called dentrites," said Prof. Shun-ichi Amari, RIKEN Brain Science Institute, Japan. He was delivering the keynote address on 'Dynamics of Learning in Neural Networks' at the 11th International Conference on Neural Information Processing organised by the Indian Statistical Institute (ISI) and Jadavpur University at the Science City recently. 

"Neurons send out electrical signals through a long, thin strand known as the axon, which splits into thousands of branches," said Amari. "Those branches are connected with other neurons through structures called synapses. The electrical signal excites other neurons via the synapse." This chain of events leads to the storage of information like odour, colour and shape - phenomenon that we call learning. 

An artificial neural network can be a real electronic circuit made of physical nodes (chips), analogues of neurons and connections, analogues of the axons and dentrites through which neurons are linked. Sometimes scientists use computers to simulate the behaviour of such a learning device. 

"The neural networks' ability to learn has a wide range of applications from finding a fault in an aircrafts' design to checking fraud in banking systems," said Dr. K. Kasturirangan, director of the National Institute of Advanced Studies, Bangalore. 

According to him, neural networks, being quick learners, can prove handy in facing unpredictable situations in planetary exploration. "Scientists are also using this device to enhance a child's learning capacity," he added. 

As neural networks operate on binary logic of the digital computers, they can't yield information that humans can perceive. To overcome this limitation, scientists merged fuzzy logic with neural networks. This has widened the scope of neural networks' applications, one example being the artificial nose. Dr. Cleber Zanchettin of the Federal University of Pernambuco, Brazil, discussed this in his speech on 'Evolving Fuzzy Neural Networks Applied to Odor Recognition'. "We have developed artificial nose that can detect and classify five gases: ethane, methane, butane, propane and carbon monoxide from the petroleum industry," said Zanchettin.