This paper reviews recent developments in deep learning and unsupervised feature learning for time-series modeling. It discusses the challenges of time-series data, such as high dimensionality, noise, non-stationarity, and temporal dependencies, and presents various models and techniques for modeling these characteristics. The paper covers models like Restricted Boltzmann Machines (RBM), Conditional RBM, Gated RBM, Auto-encoder, Recurrent Neural Network (RNN), Deep Learning, Convolution and Pooling, Temporal Coherence, Hidden Markov Model, and others. It also discusses applications of these models in various time-series problems, including video processing, stock market prediction, speech recognition, music recognition, and motion capture data. The paper highlights the importance of learning features from unlabeled data and the potential of deep learning models to capture complex temporal relationships in time-series data. It concludes that deep learning methods, particularly those that incorporate temporal coherence and feature learning, offer promising approaches for modeling and analyzing time-series data.This paper reviews recent developments in deep learning and unsupervised feature learning for time-series modeling. It discusses the challenges of time-series data, such as high dimensionality, noise, non-stationarity, and temporal dependencies, and presents various models and techniques for modeling these characteristics. The paper covers models like Restricted Boltzmann Machines (RBM), Conditional RBM, Gated RBM, Auto-encoder, Recurrent Neural Network (RNN), Deep Learning, Convolution and Pooling, Temporal Coherence, Hidden Markov Model, and others. It also discusses applications of these models in various time-series problems, including video processing, stock market prediction, speech recognition, music recognition, and motion capture data. The paper highlights the importance of learning features from unlabeled data and the potential of deep learning models to capture complex temporal relationships in time-series data. It concludes that deep learning methods, particularly those that incorporate temporal coherence and feature learning, offer promising approaches for modeling and analyzing time-series data.