networks, networks everywhere!

Eric J. Ma (MIT)

slides are available online

about myself

  • doctoral candidate, MIT biological engineering
  • self-taught pythonista
  • using networks to solve problems in infectious disease ecology, evolution & biochemistry

outline

  1. what are networks?
  2. example 1: recommendation systems
  3. example 2: panama papers
  4. example 3: influenza ecology & evolution
  5. example 4: neural networks on networks

what are networks

networks, a.k.a. graphs, are composed of nodes (circles) and edges (lines)
networks, a.k.a. graphs, are composed of nodes (circles) and edges (lines)

example 1: recommendation systems

if A is connected to B and C, but B and C are not connected, then maybe they should be!
if A is connected to B and C, but B and C are not connected, then maybe they should be!

example 1: recommendation systems

if A and B share overlapping interests, then maybe some of B's interests can be recommended to A.
if A and B share overlapping interests, then maybe some of B's interests can be recommended to A.1

example 2: panama papers

graph databases were used to show how the rich hide their money.
graph databases were used to show how the rich hide their money.2

example 3: influenza ecology and evolution

for influenza, gene shuffling probably helps in host switching.
for influenza, gene shuffling probably helps in host switching.3

example 4: neural networks on networks

graph convolutions let us do machine learning on graph-structured data.
graph convolutions let us do machine learning on graph-structured data.4

visualize networks rationally

move away from the hairball!
move away from the hairball!

visualize networks rationally

rational network visualizations prioritize placement of nodes
rational network visualizations prioritize placement of nodes

conclusions

  • think relationally
  • networks can be used creatively to solve all sorts of problems

keep in touch


  1. Collaborative filtering

  2. International Consortium of Investigative Journalists (ICIJ) and Neo4j unravel the panama papers.

  3. Reticulate evolution is favoured in influenza niche switching.

  4. Convolutional Networks on Graphs for Learning Molecular Fingerprints