LESSON 4 - Neural computation Flashcards
At a basic level, what do neurons do?
Neurons detect the presence of certain conditions and signal whatever they detect. They perceive signals from sensory receptors or other neurons, forming layers organized hierarchically.
How does the receptive field of a neuron in the visual cortex relate to its activation?
The receptive field is the region in sensory space that can activate a neuron. If a stimulus covers both center and non-center regions, they may cancel out each other, resulting in no response.
What did Hubel and Wiesel discover about neurons in the visual cortex?
Hubel and Wiesel found that neurons in the visual cortex are tuned to oriented parts of edges. Neurons respond to specific orientations, and their tuning to properties is not predetermined but is a result of connectivity to other neurons.
How do artificial neural networks (ANNs) relate to biological systems?
ANNs are inspired by biological systems but aim to capture basic principles rather than reproducing all the complexity. Even a single neuron in ANNs is a complex system, working dynamically through electro-chemical processes.
What is the main distinction between the AI perspective and the cognitive science perspective in the context of artificial neural networks?
The AI perspective sees ANNs as powerful machine learning methods for solving tasks, while the cognitive science perspective uses ANNs to simulate cognitive abilities and behavior for study rather than practical application.
When was the first mathematical formalization of artificial neurons, and who proposed the idea that learning changes synaptic connections?
The first mathematical formalization was in 1943, and Donald Hebb in 1949 proposed that learning changes the strength of connections between neurons, known as synapses.
What is the significance of the connectionism movement in the 1980s?
The connectionism movement in the 1980s, particularly the PDP group in 1986, played a significant role in influencing both artificial neural networks and psychology.
What is the purpose of the activation function in a formal neuron?
The activation function, such as the sigmoid function, transforms the integrated input signals into an output between 0 and 1. It determines the neuron’s activity, and certain functions like sigmoid ensure the output does not reach 1, no matter how much further stimulated.
What is the role of hierarchy in neural networks concerning information abstraction?
Through hierarchical organization, neural networks can encode increasingly abstract information. The type of information encoded becomes more abstract as you move up the hierarchy.
How did Hubel and Wiesel contribute to understanding neural responses in the visual cortex?
Hubel and Wiesel, through experiments on cats’ visual cortex, discovered that neurons are tuned to oriented parts of edges. The tuning is not inherent but results from connectivity to other neurons.
What is the significance of conditional independence in the context of neurons’ connectivity?
Conditional independence allows exclusion of irrelevant connections, simplifying neural network graphs. Neurons that are only connected to a subset of others can directly influence specific values, making the graph more organized.
How do artificial neural networks differ from traditional computers in terms of processing?
Unlike computers with a general purpose, artificial neural networks have specialized processing. Neurons accumulate evidence for a specific condition and communicate the result to other neurons, allowing for complex computations in a network.
What distinguishes the Bayesian perspective in probability theory?
In the Bayesian perspective, the focus is on establishing a degree of belief that can change over time based on observations. This perspective allows for adapting opinions and probabilities based on new evidence, like the concept of a “Black Swan.”
How do artificial neural networks inherit key properties from biological systems?
Artificial neural networks inherit properties like learning from experience and adapting to novel contexts, allowing them to perform tasks without explicit knowledge of the problem, solely through learning and experience.
efine machine learning and differentiate it from artificial neural networks.
Machine learning refers to any method for learning from data, and artificial neural networks are one method within machine learning. Machine learning encompasses various approaches, and choosing the right one depends on the problem at hand.
