Skip to content

Attending the Data Science Institute seminar at Imperial College


impdsi gautamshroff

impdsi gautamshroff

DSI Talk with Dr Gautam Shroff 19/02/2015 – NOTICE: These are only notes and they may not make much sense out of context… 

Current business themes – Digital imagination – Simplification – Governance – Suistainability

Digital re-imagination

From = Data “raw material” to = Delivery “means”

  • Internet of things
  • Social
  • Big Data
  • Mobility, Cloud
  • Buzz word of “analytics”

What does big data mean? Not that its a lot of data but its width -> High dimensional spaces.

Things to take into account:

  • Data width v Data length

  • Real-time response v Strategic response

Simulation is very important although it is not traditionally in the arena of predictive analytics.

The typical path:


  • analise historical customer transactions
  • analyse historical offers
  • predict each offer type probability


  • for a desired cost-benefit ratio find unique customer-offer strategy

But that is not enough. In practice there are a number of external factors to take into account. So you try or simulate

The typical flow #predictive #analytics is prediction & optimisation. But the real challenge are external factors. So try or #simulate #DSI

Unfortunately trying is not a viable option for a number of businesses out there!

Agent-based modelling and Casual Analysis are not purely statistics, you need assumptions to set them up!

Detecting abnormal behaviour using deep learning. Look at Office power consumption:

EXAMPLE: Enterprise contextual intelligence: detecting specific event from twitter

  • Supplier intelligence
    • Detecting events that could adversely impact a supplier (fire, strike, flood, etc) and pushing these based on conversation/transactional context
  • Context-aware information advertisements
    • detecting nature user-activity; discerning need and pushing top relevant internal/external social network updates
  • Correlating unstructured events with structured data
    • detecting significant news events using entity-specific indicators and correlating them