Toolkit for Multivariate Analysis (TMVA) is a multi-purpose machine learning toolkit integrated into the ROOT scientific software framework, used in many particle physics data analyses and applications. The following are also areas of interest with impactful applications to particle physics.
Task ideas and expected results
- Deep Q-Learning: a deep reinforcement learning technique combining the standard Fully-Connected Networks using a biologically inspired technique called experience replay https://arxiv.org/abs/1312.5602.
- Optimization module: Momentum-based, Adam, RMSProp. The optimizers can be developed separately, because they only need the labels and the objective function of the model.
- Deep Learning Module Input: low-level implementation of a 3D (or possible higher order) tensor, which will manage the memory better. Also, a better interface to read a ROOT file and supply it directly to the Deep Learning Network.
- Gaussian Processes: this is a classical, but powerful method for variational inference.
- Unsupervised learning:
- deep auto-encoders, restricted boltzmann machines (RBMs)
- Deep learning:
- LSTMs, complex-valued neural networks
Strong C++ skills, good understanding of machine learning algorithms