By learning suitable and well-performing dialogue behaviour in statistical spoken dialogue systems has been in the focus of research for many years. While most work which is based on reinforcement learning employs an objective measure like task success for modelling the reward signal, we proposed to use a reward based on user satisfaction. We evaluated different methods such as Linear Regression, SVM, RNN, etc. to compare the pros and cons of each method. The model will later be used to assist the training of other components of conversational AI such as dialogue manager, etc. to be more robust and converse in a more satisfactory reliable ways.

Unexpected IT equipment failures are a major cause of downtime in data centers. To prevent this, we aim to predict equipment failures a week ahead of time. For each device, we collect data on utilization and event logs. This data holds valuable information, but it's challenging to analyze due to its variable length, inconsistent sampling, and missing entries. We developed a model using Long Short-Term Memory (LSTM) cells within a Recurrent Neural Network (RNN) architecture to predict device failures based on this data. To address the data imbalance in our dataset, we implemented data augmentation and a weighted loss function. We explored bi-directional LSTMs and sequence labeling techniques to maximize the accuracy of our predictions. Finally, we used grid search to optimize the model's hyperparameters. Compared to a baseline random forest classifier, our model achieved a significant improvement of 24% in F1 score.