Ken,

> My own view is that some time should be spent on non-neural network
> techniques to give "a much broader perspective on machine learning".

Great.

> While other forms of machine learning may be simpler to do and
> understand they are much more limited.

Why do you believe they're more limited? Any supervised learning
method can apply to "tabular" data and there have even been some
impressive results with other approaches on image processing.

> Neural network research and real-world applications greatly outnumber
> other techniques these days.

Yes, but one major reason for that imbalance is that so few junior
researchers and developers have been exposed to the alternatives.
Don't we want them to have access to all available tools?

> And introducing students to a variety of programs takes lots of time.

There's always a tradeoff between offering broad coverage to students
and drilling down into details of a specific approach. Which do you
think is more appropriate for an introduction to machine learning?

Best, -Pat