Adrian Roman

This could be done for EMNIST or something else. Train an autoencoder (encoder + decoder) to learn an efficient encoding/features representation in an unsupervised manner, then throw away the decoder,...

A little more complex than the simple RNN but doable... maybe sometime.

A basic RNN is fairly simple to implement, see for example https://gist.github.com/karpathy/d4dee566867f8291f086 so why not?

See for example https://arxiv.org/abs/1607.06450

ZWave plus mandatory classes

3

In ZWavePlusDeviceClass.java, ZWavePlusDeviceClass, ZWavePlusDeviceType enum: They appear to be in order, but at about line 81 ``` SEISMIC_INTENSITY_SENSOR(0x0d19, DEVICE_RESET_LOCALLY, ZWAVE_PLUS_INFO, ASSOCIATION_GROUP_INFO, MANUFACTURER_SPECIFIC, POWERLEVEL, ASSOCIATION, VERSION, SENSOR_BINARY, SENSOR_MULTILEVEL), SEISMIC_MAGNITUDE_SENSOR(0x0d1a, DEVICE_RESET_LOCALLY, ZWAVE_PLUS_INFO,...

Local pseudopotentials would be very easy (see the python repository for an implementation), the ones available are non-local ones and make things more complex...

Probably I won't add it, but VWNExchCor has LSDA implemented already.

The single 'quantum dot' could be extended to two of them in an 'artificial molecule' or even a molecule could be computed (the code is already available https://github.com/aromanro/DFTQuantumDot/blob/master/DFT/Molecule.h but no...

https://en.wikipedia.org/wiki/Wolff_algorithm

It's easy to implement, the UI changes would be more difficult than the actual model