Optimizing your talk to an academic audience: a mathematical approach

This one came to me while musing about a number of things on a bus trip. Engineers love to model things, so I figured why not apply these techniques to something we engineering researchers have to do quite a bit: giving a talk to an academic audience.

Let’s make some reasonable assumptions. First, any talk presents a number of ideas and touches on a number of topics. So we can model the talk as a time series , and over any time window we can do a transformation () to find the topic content (analogous to the Fourier analysis of a signal).

Next, let’s assume the average academic in the audience is interested in a number of topics or ideas and only pays attention when his or her interest is peaked by this topic being present in that talk at any point in time. Of course if the speaker moves on to another topic which is not of interest, then the academic will lose interest over time, until a topic he or she is interested in finds its way into the talk again. That means an academic can be modeled as having a front-end of a number of ‘band pass’ topic filters () and a back-end of a finite-impulse response (FIR) filter () that feeds his or her attentiveness ().

The ‘band-pass’ filters remove from any topics that are not of interest to the academic producing an output which will be zero if the talk at that point in time does not contain topics of interest to the academic and non-zero if it does. The FIR filter models at the time to get the academic’s attention once the speaker touches on a topic of interest and the time to lose that attention once the speaker hits a topic not of interest. The FIR filter is an averaging filter that weights the current and past inputs to it. Its input is and its output is high when the current part of the talk contains a topic of interest and the previous parts contained topics of interest as well and low when either the current part or previous parts that the filter ‘remembers’ contain topics not of interest.