Software
machine learning:
Neural Modules (NeMo) (2019 - ...): a toolkit for conversational AI. [ Source ]
PyTorchPipe (PTP) (2019 - ...): is a component-oriented framework facilitating development of computational multi-modal pipelines and comparison of diverse neural network-based models. [ Source ]
MI-Prometheus ([M]achine [I]ntelligence - Prometheus) (2018 - ...): MI Prometheus is an open source Python library, built on top of PyTorch, that enables reproducible Machine Learning research. [ Source | Docs ]
Kornuta, T., Marois, V., McAvoy, R.L., Bouhadjar, Y., Asseman, A., Albouy, V., Jayram, T.S. and Ozcan, A.S. (2018). Accelerating Machine Learning Research with MI-Prometheus. NeurIPS 2018 workshop on Machine Learning Open Source Software (MLOSS) [ OpenReview | .pdf ]
MIC ([M]achine [I]ntelligence [C]ore) (2015 - ...): A (distributed) framework for machine intelligence. The project distribution (in terms of distribution into many, depending on each other, repositories) targets the need for realization of many diverse projects, supposed to share the same algorithms at their core.
MI-Toolchain: A subproject of MIC, being the real "core" of the framework. Contains tools required for development of MIC-based applications, such as configuration management, properties, loggers, application state, event handlers etc. [ Source | Docs ]
MI-Algorithms: A sub-project of MIC framework. The project contains core types (sample, batch, matrix tensor) and tools (MNIST/CIFAR/STL/RawText importers, encoders etc.) useful when training different models and working with different problems. [ Source | Docs ]
MI-Visualization: A sub-project of MIC framework. Contains tools for building applications with OpenGL-based visualization of data, results of execution etc. [ Source | Docs ]
MI-Neural-Nets: A sub-project of MIC framework. The repository contains algorithms and applications related to multi-layer (deep) feed-forward (for now) neural nets. In particular, it contains implementations of diverse layers (fully connected, sparse, hebbian, convolution, pooling etc.), activation functions (ELU, ReLU, Softmax etc.) and optimizers (SGD, SGD with momentum, AdaGrad, AdaDelta, Adam, Hebb's update rule etc.) with optimization landscapes (for testing and comparison of different optimizers). Plus a plethora of unit tests (e.g. for forward and backward passes through fully connected, convolution). [ Source | Docs ]
MI-Reinforcement-Learning: A sub-project of MIC framework. The repository contains solutions and applications related to (deep) reinforcement learning. In particular, it contains several classical problems (N-armed bandits, several variations of Gridworld), POMDP environments (Gridworld, Maze of Digits, MNIST digit) and algorithms (from simple Value Iteartion and Q-learning to DQN with Experience Replay). [ Source | Docs ]
Kornuta, T. & Rocki, K.(2017). Utilization of deep reinforcement learning for saccadic-based object visual search. In International Conference Automation (pp. 565-574). Springer, Cham. [ Springer | ArXiv ]
computer vision:
DisCODe ([Dis]tributed [C]omponent-[O]riented [D]ata Proc[e]ssing): DisCODe is a framework facilitating development of sensory data processing algorithms. DisCODE is written mostly in C++ (in an objective manner) and composing of a components library with patterns for their usage. Those patterns impose general implementation of diverse, multistage data processing algorithms. (2010 - ...)
FraDIA ([Fra]mework for [D]igital [I]mages [A]nalysis): FraDIA is a framework, written mostly in C++ and composing of a components library with patterns for their usage. Those patterns impose general implementation method of image processing and analysis algorithms as well as construction of user interface, which might be required for algorithms finetuning and testing. FraDIA can work as a stand–alone application, as well as a vision subsystem of the MRROC++ based robotic controllers. (2005 - 2010)
Kornuta, T. Bem, T., and Winiarski, T. (2011). Utilization of the FraDIA for development of robotic vision subsystems on the example of checkers' playing robot. Machine GRAPHICS & VISION, 4:495-520. [ .pdf ]