There is no real utility in taking the position that you won’t do model selection via validation splits, although, know there is a meta-purpose to bringing a blowtorch to this concept…

A common application of gaussian mixture models is in property pricing. It’s used as an implementation of a “Hedonic Pricing” which involves modelling the price using variables categorised into 3 types; structural (intrinsic properties of the property), locality (region, state etc.) and environmental (air quality, proximity to parks etc.) the latter 2 types genrally have noise in their measurement - so are modelled as ‘random effects’. it’s this that makes a mixture model needed, where conventional multiple regression would only have errors as randomly distributed, and predictors fixed

In this 5-min read I’ll try to explain how convergence can be slower for LSTM neural networks where memory state is maintained from one mini-batch to the next. An example of this problem can be found in my project about a microsleep detector

It’s much more costly to the end user for the autonomous vehcile to not take over the steering when eyes are closed, than to take over steering when eyes are open.

Autonomous vehicles are no longer reserved for the realm of science fiction, but some AI experts say the complexities of AI-driven cars in public make the possibility of never needing to drive a far off future. Perhaps to bridge the gap are technologies that greatly reduce the labour involved in driving, while still reducing the risk.

Produce a detector that distinguisihes when a car’s driver has eyes closed. This detector will connect to an autopilot system for the car temporarily until driver regains control, otherwise safely brings the car to a halt on the side of the road. Detector will have access to EEG signals connected to drivers head.

Explaining the concept of ‘Attention’ in Natural Language Processing Models by removing part of the memory function of a Recurrent Neural Network Encoder-Decoder

See the LSTM building & training algorithm I developed as a learning excercise on my Github