Week 2

Information Theory (Shannon, 1948)

• Figured out how to quantify / measure information (bit)

• Deduced channel capacity → which information must be compressed and by how much in order to be transmitted

• Information entropy (how uncertain / unpredictable a message is)
  = the more random and diverse information is, the less it can be compressed

Efficient encoding in sensory systems

Sparse representation: to conserve energy the brain aims to encode as much information using as few neurons as possible
→ optimised for natural environment, patterns we encounter frequently

1. Cones are not evenly spaced across light wavelengths

   • Close red-green spacing increases sensitivity

   • Optimised for environment, colours in nature are not random
     

2. Receptor density across the retina

• FOVEA: high cone density

  • ‘Simple retinal circuit’ = 1:1 photoreceptor to neuron ratio

  • High acuity, good for fine detail

  • Colour vision limited to our focal point, our eyes move to analyse detail and colour

• PERIPHERY: high rod density

  • ‘Convergent retinal circuit’ = 7:1 photoreceptor to neuron ratio

  • High sensitivity to light, poor detail

How does prior knowledge influence percpetual interpretation?

‘Top-down knowledge’, using something you already know to solve a novel or ambiguous situation

• Bayesian framework: what is the probability of something given 2 sources of information

• Prior probability (previous knowledge) + New evidence (information actively coming in)

• The brain uses this to solve ambiguity in perception (fill in gaps, ‘clean-up’ representations, assume or apply based on previous experience)

3 levels of analysis for cognitive and neural functions (Marr, 1982)

1. Computational level
   → what is the goal of the system being studied?

2. Representation & algorithm
   → what is the format of this information during input vs processing vs output?

3. Hardware implementation
   → what is the physical hardware (biological vs mechanical) used to accomplish the task? (implement the algorithm)

How do receptive fields and neural circuits construct complex features from sensory input?

i.e., edges, motion

Cortical maps and modular organisation in sensory and higher-order processing

Behavioural, fMRI and MEG data in methods to measure the brain’s representational structures