The Turing++ Problems

The challenge is to develop computational models that answer questions about images and videos such as.

What is there / Who is there / What is the person doing

and eventually more difficult Questions such as...

Who is doing what to whom? /What happens next at the computational, psychophysical, and neural levels

So we can say, that to understand human intelligence we must

1. Understand what we compute

2. How what we compute develops

3. How amplified by social interactions

4. How implemented in neural tissues

Orientation tuning

A neuron receives information from a dendrite Integrates that information and decides whether to fire a spike or not.

Now what happens in Neuron or how does it decide to fire? There are different models describing them.

1. Filter operations

2. Integrate and Fire circuits

   This is basically like an RC (Resistor-Capicator) circuit, small charges are input signals that accumulate and then fire when there is enough charge.

   

3. Hogdkin-Huxley units

4. MultiCompartmental model

5. Spines Channel

Neural Circuitry

When we connect multiple Neurons together This is known as a Network, The Picture below tries to show each neuron in the form of a circle

The signal starts from bottom to top, so we could say the bottom layer is the input layer, and the top layer is the output layer, and the intermediate layers are called hidden layers. ( Similar to what we know in machine learning)

Feed-Forward connection: Information flowing in one direction (typically input to output)

Feedback connection: Information flowing in the opposite direction

Recurrent connection: horizontal connections within a particular layer

For some Detail on how Image is processed in Brain see Image in Human

Feedback Connections

It has been seen that the Primary Visual cortex (v1) has more connection coming from the Secondary Visual cortex (V2) rather than a connection from V1 itself. (feedback) rather than (feedforward)

so why are there so many feedback connections?

Computational roles of feedback signals.

1. Fundamental computation in V1

2. Visual search

3. Pattern completion

One of the things that Hubel and Wiesel did was record readings in V1 and showed that there are neurons that show orientation tuning.

eg: In the below image they showed that in a cat brain, some neuron was firing very vigorously when this orientation was shown and had no effect on any other neuron.

This is now what we define by Gaabor functions, many deep-convolution start with filters that resemble these filter

$$g(x, y) = exp(-(x'^2 + γ^2 * y'^2) / (2 * σ^2)) * cos(2 * π * x' / λ + φ)$$