Earlier this week, I worked my way through Giora Simchoni’s presentation on Deep Visual Inference: Teaching Computers To See Rather Than Calculate Correlation (link).
One of Giora’s slides was:
As well as telling me about the learning journey I need to make to understand Giora’s processes and outcomes, George Box popped into my head as I was reading.
Back in 1979, George wrote about robustness in scientific model building (link). In his paper, George observed:
The process of robustification is best carried out … by suitably elaborating the model and then using classical estimation procedures.
George defines robustness as “the property of a procedure which renders the answers it gives departures of a kind that occur in practice, from ideal assumptions”.
George talks about parsimony and iteration in model building and presents this schematic (1979:3) about the process of model building:
I do think there is an explicit link between Giora and George. Reading Giora first meant I could return to George. The challenges Giora discusses have their place in George’s parsimony and iteration. These do include conversations about non-linearity and sample sizes. I took them to be very important too in how we visualise data.
To this end I really enjoyed Giora’s final slide:
I think George’s response might be to discuss ‘departures‘.