1.   Why is this a computer science course?
The brain is an incredible computational device which has a completely different computational structure and procedure from modern computers.  Computer science can learn much about solving difficult problems by studying ways in which the brain solves some of its computational problems.

Differences between brains and computers:

- Brains are much slower than computers (signals measured in milliseconds)
- Brains have much higher connectivity. Each neuron is connected to up to 10,000 other neurons.
- Brain computation is massively parallel. Neurons fire together rather than strictly sequentially.
- Brains are highly evolved and are trained, not designed and programmed.
- Brains do not always give the same action in response to the same circumstances, unlike computers, which are expected to give the same response for the same circumstances.
- Brain signals are gradable and probabilistic, not digital, deterministic.

In addition, computer scientists are interested finding more natural ways for humans to communicate with computers. Speech and Natural Language interfaces are among the most promising current approaches.  To meet the future requirements, language understanding systems will need to incorporate deep semantic understanding that appears to require embodiment. More immediately, all user interfaces are based on metaphors (e.g., the desktop) and there is active work on using metaphor theory in design.

Finally,  the connectionist systems we will be discussing are themselves important contributions to the theory and practice of massively parallel computation.

2.   Basic brain structures
Note: Many of the slides used as illustrations in class are part of the Scientific American articles assigned for this weeks readings.

A myth: we don't know anything substantial about the brain.

Today's lecture dispels this myth, but also illustrates how very complex the brain is. The handout given in class separates the brain study into 7 levels of detail.  The first three levels, behavior, systems and local circuits, will be the main focus of the course.

The neuron:
See today's reading for a detailed description of typical neuron structure.

The neural signal is in part electrical and in part chemical. For one neuron to send signals to up to 10,000 others,  there needs to be a source of energy.  As described in the reader, axons have active membranes that propogate signals by a process involving voltage-gated channels and pumping mechanisms to restore the membrane potential difference.

A detailed description of the exchange of potassium and sodium particles which propagates the signal for uninsulated axons is in the readings for this class.  Many axons, however, have a myelin sheath which wraps around the axon, insulating it. Breaks in the sheath, called Nodes of Ranvier allow the chemical exchange of potassium and sodium which propagates the signal. Between nodes, however, the signal is transmitted by electrical current.

Synapses:
Synapses are also explained in detail in the reading. Synapses may be on dendrite receiving spines (small spines protruding from dendrites), on the main dendritic branch or on the cell body itself.  Synapses on the cell body are usually inhibitory connections, which prevent the neuron from firing. The parts of sending and receiving neurons which make up synapses are held in place by surrounding glial cells.
 

Brief Overview of some motor systems:

More is known about the motor and visual systems of the brain than about the language systems because (1) these systems in humans are similar to those in other animals which can be closely studied and (2) motor and sensory systems seem to be
more localized than language.

Systems are controlled on several levels and through a top-down system.  For instance, a neuron may fire for a particular mode of behavior, but will not fire in another mode. The sensory information to a system may also excite or inhibit firing, so that the output of the system depends on external, higher-level input.

Motor control systems use the basic computational strategies of mutual inhibition and mutual excitation to coordinate behaviors, such as gait. These strategies are also basic for language behaviors such as speech recognition and ambiguous sentences. Basic motor control systems are also the basis for neural-computational theory of linguistic aspect and metaphor.  These connections will be discussed in detail in a later lecture.

For those interested in learning more details about neurobiology, consult Introduction to Neurobiology by Heinrich Reichert.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

References:  Reader:1, 2, 3; Regier: ch 1.