UC BERKELEY COGSCI 1 MIDTERM
SUMMER 2026/2027 ACTUAL
QUESTIONS WITH VERIFIED ANSWERS.
Single layer networks - correct answer-Mapping function: Maps
a set of items (domain) to a set of items (range).
Function: Each item in the domain has one (and only one)
value in the range.
Multilayer networks - correct answer-Units arranged in layers:
Input layer
Hidden layer(s)
Output layer
Hidden layer - correct answer-The second layer of a three-layer
network where the input layer sends its signals, performs
intermediary processing
Feedforward - correct answer-activation spreads forward
through the network - there is no spread between units in a
layer or backwards
,Backpropagation algorithm - correct answer-a learning
algorithm in multilayer neural networks in which error is spread
backwards through the network from the output units to the
hidden units, allowing the network to modify the weights of the
units in the hidden layers
Local algorithms - correct answer-individual unit's weight
changes directly as a function of the inputs to and outputs from
that unit. It's like Hebbian learning. This algorithm is more
biologically plausible than backpropagation.
Competitive networks - correct answer-each out put fires in
response to a particular set of input patterns. The unit that fires
the most will win the competition. These are good for
classification tasks.
Unsupervised learning - correct answer-like Hebbian learning -
no feedback required - it's like classical conditioning
Supervised learning - correct answer-A type of machine
learning where algorithms are trained by providing feedback
about incorrect solutions - like size of the error.
,Epsilon (squiggly e) - correct answer-learning rate - it's a
constant
Gradient descent learning - correct answer-learning routine or
technique applied to neural networks in backpropagation -
learning can stop when the slope is zero
Distributed networks - correct answer-characterized by many
different hubs and links, which allows continuous
communication even if some connections stop working
Dynamical system - correct answer-a system made up of many
interacting elements that changes and evolves over time
It is a system that evolves over time in a rule governed way. It
can be studied using tools of dynamical modeling.
State space - correct answer-all possible states the system can
be in (like where a pendulum could be located)
, Four features of a computational governor - correct answer-1.
Representational (represents parameters of the problem)
2. Computational - can manipulate representations
3. Sequential (works in a step by step manner)
4. Decomposable (homuncular - made up of distinct
subsystems - each responsible for a particular subtask)
Initial conditions - correct answer-where the system starts
Watt governor - correct answer-regulate the speed in a steam
engine to make it uniform - done in a mechanical way not a
computational way
Four features of Watt governor (by Van Gelder) - correct
answer-Dynamical system
Time sensitive
Coupling (flywheel touches the throttle valve)
Attractor dynamics - will reach an equilibrium arm angle
Basins of attraction - correct answer-regions of state space
where trajectories converge (like a pendulum being stationary
at its equilibrium point)
SUMMER 2026/2027 ACTUAL
QUESTIONS WITH VERIFIED ANSWERS.
Single layer networks - correct answer-Mapping function: Maps
a set of items (domain) to a set of items (range).
Function: Each item in the domain has one (and only one)
value in the range.
Multilayer networks - correct answer-Units arranged in layers:
Input layer
Hidden layer(s)
Output layer
Hidden layer - correct answer-The second layer of a three-layer
network where the input layer sends its signals, performs
intermediary processing
Feedforward - correct answer-activation spreads forward
through the network - there is no spread between units in a
layer or backwards
,Backpropagation algorithm - correct answer-a learning
algorithm in multilayer neural networks in which error is spread
backwards through the network from the output units to the
hidden units, allowing the network to modify the weights of the
units in the hidden layers
Local algorithms - correct answer-individual unit's weight
changes directly as a function of the inputs to and outputs from
that unit. It's like Hebbian learning. This algorithm is more
biologically plausible than backpropagation.
Competitive networks - correct answer-each out put fires in
response to a particular set of input patterns. The unit that fires
the most will win the competition. These are good for
classification tasks.
Unsupervised learning - correct answer-like Hebbian learning -
no feedback required - it's like classical conditioning
Supervised learning - correct answer-A type of machine
learning where algorithms are trained by providing feedback
about incorrect solutions - like size of the error.
,Epsilon (squiggly e) - correct answer-learning rate - it's a
constant
Gradient descent learning - correct answer-learning routine or
technique applied to neural networks in backpropagation -
learning can stop when the slope is zero
Distributed networks - correct answer-characterized by many
different hubs and links, which allows continuous
communication even if some connections stop working
Dynamical system - correct answer-a system made up of many
interacting elements that changes and evolves over time
It is a system that evolves over time in a rule governed way. It
can be studied using tools of dynamical modeling.
State space - correct answer-all possible states the system can
be in (like where a pendulum could be located)
, Four features of a computational governor - correct answer-1.
Representational (represents parameters of the problem)
2. Computational - can manipulate representations
3. Sequential (works in a step by step manner)
4. Decomposable (homuncular - made up of distinct
subsystems - each responsible for a particular subtask)
Initial conditions - correct answer-where the system starts
Watt governor - correct answer-regulate the speed in a steam
engine to make it uniform - done in a mechanical way not a
computational way
Four features of Watt governor (by Van Gelder) - correct
answer-Dynamical system
Time sensitive
Coupling (flywheel touches the throttle valve)
Attractor dynamics - will reach an equilibrium arm angle
Basins of attraction - correct answer-regions of state space
where trajectories converge (like a pendulum being stationary
at its equilibrium point)
