reading.md

Reading for Machine Learning

http://neuralnetworksanddeeplearning.com/

• Chapter 1 (Neural network introduction) ✅
• Chapter 2 (Backpropagation) ✅
• ...

Reading for Statistical Models / MVSs

Coursera Machine Learning Course

• Week 1 ✅
• Week 2 ✅
• Week 3
• Week 4
• Week 5
• Week 6
• Week 7
• Week 8
• Week 9
• Week 10
• Week 11

Conklin, Witten: Multiple viewpoint systems for music prediction (1995)

Pearce, Wiggins: Towards a Framework for the Evaluation of Machine Compositions (2001) ✅

• Core idea: to make our work more scientific, we need to present and provide evidence for refutable claims.
• E.g. "People cannot distinguish the machine composed music from human composed music"
• "Framework" for evaluation:
  • notion of a machine-learned "critic", which ensures style adherance
  • evaluate by asking human subjects to distinguish compositions from data set vs generated ones
  • => machine compositions evaluated objectively within a closed system which does not consider aesthetic merit.
• Many subtleties discussed in Section 6 onwards; subjects exhibited bias towards classifying as machine generated

Conklin: Music Generation from Statistical Models (2003) ✅

• Overview of random walk versus Viterbi decoding vs sampling.
• Random walk often fails to lead to overall high probabiltiy solutions.

## Whorley: Construction and Evaluation of Statistical Models of Melody and Harmony (PhD thesis, 2013)

• Explores multiple viewpoint frameworks as statistical models of melody and harmony.
• Multiple viewpoint framework was originated by Conklin and Cleary (1988)

Whorley: Music Generation from Statical Models of Harmony (2016)

Pearce: The construction and evaluation of statistical models of melodic structure in music perception and composition (PhD thesis, 2005)

• Shows multiple viewpoint systems being used successfully