Visual Interpretation of Recurrent Neural Network on Multi-dimensional Time-series Forecast

Recent attempts at utilizing visual analytics to interpret Recurrent Neural Networks (RNNs) mainly focus on natural language processing (NLP) tasks that take symbolic sequences as input. However, many real-world problems like environment pollution forecasting apply RNNs on sequences of multi-dimensional data where each dimension represents an individual feature with semantic meaning such as PM2.5 and SO2. RNN interpretation on multi-dimensional sequences is challenging as users need to analyze what features are important at different time steps to better understand model behavior and gain trust in prediction. This requires effective and scalable visualization methods to reveal the complex many-to-many relations between hidden units and features. In this work, we propose a visual analytics system to interpret RNNs on multi-dimensional time-series forecasts. Specifically, to provide an overview to reveal the model mechanism, we propose a technique to estimate the hidden unit response by measuring how different feature selections affect the hidden unit output distribution. We then cluster the hidden units and features based on the response embedding vectors. Finally, we propose a visual analytics system which allows users to visually explore the model behavior from the global and individual levels. We demonstrate the effectiveness of our approach with case studies using air pollutant forecast applications.

[1]  Scott Lundberg,et al.  A Unified Approach to Interpreting Model Predictions , 2017, NIPS.

[2]  Judith C. Chow,et al.  Characteristics of carbonaceous aerosol in Pearl River Delta Region, China during 2001 winter period , 2003 .

[3]  Joachim Diederich,et al.  Survey and critique of techniques for extracting rules from trained artificial neural networks , 1995, Knowl. Based Syst..

[4]  Alexis K.H. Lau,et al.  Seasonal characteristics and regional transport of PM2.5 in Hong Kong , 2005 .

[5]  Zhen Li,et al.  Understanding Hidden Memories of Recurrent Neural Networks , 2017, 2017 IEEE Conference on Visual Analytics Science and Technology (VAST).

[6]  Xinlei Chen,et al.  Visualizing and Understanding Neural Models in NLP , 2015, NAACL.

[7]  P. Rousseeuw.  Silhouettes: a graphical aid to the interpretation and validation of cluster analysis , 1987 .

[8]  Xiaogang Wang,et al.  Deeply learned face representations are sparse, selective, and robust , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[9]  Hao Yang,et al.  Applications of Recurrent Neural Networks in Environmental Factor Forecasting: A Review , 2018, Neural Computation.

[10]  Fei-Fei Li,et al.  Visualizing and Understanding Recurrent Networks , 2015, ArXiv.

[11]  Inderjit S. Dhillon,et al.  Co-clustering documents and words using bipartite spectral graph partitioning , 2001, KDD '01.

[12]  Jude W. Shavlik,et al.  in Advances in Neural Information Processing , 1996 .

[13]  Alexander M. Rush,et al.  LSTMVis: A Tool for Visual Analysis of Hidden State Dynamics in Recurrent Neural Networks , 2016, IEEE Transactions on Visualization and Computer Graphics.

[14]  J. Friedman.  Greedy function approximation: A gradient boosting machine. , 2001 .

[15]  Sepp Hochreiter,et al.  Untersuchungen zu dynamischen neuronalen Netzen , 1991 .

[16]  Takio Murakami,et al.  Winter Monsoonal Surges over East and Southeast Asia1 , 1979 .

[17]  Alexander M. Rush,et al.  Seq2seq-Vis: A Visual Debugging Tool for Sequence-to-Sequence Models , 2018, IEEE Transactions on Visualization and Computer Graphics.

[18]  Elmar Eisemann,et al.  Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types , 2017, Nature Communications.

[19]  Dit-Yan Yeung,et al.  Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting , 2015, NIPS.

[20]  Zhen Li,et al.  Towards Better Analysis of Deep Convolutional Neural Networks , 2016, IEEE Transactions on Visualization and Computer Graphics.

[21]  Ming Li,et al.  Forecasting Fine-Grained Air Quality Based on Big Data , 2015, KDD.

[22]  Jimeng Sun,et al.  RetainVis: Visual Analytics with Interpretable and Interactive Recurrent Neural Networks on Electronic Medical Records , 2018, IEEE Transactions on Visualization and Computer Graphics.

[23]  Zachary C. Lipton,et al.  The Doctor Just Won't Accept That! , 2017, 1711.08037.

[24]  Elmar Eisemann,et al.  Hierarchical Stochastic Neighbor Embedding , 2016, Comput. Graph. Forum.

[25]  Carlos Guestrin,et al.  "Why Should I Trust You?": Explaining the Predictions of Any Classifier , 2016, ArXiv.

[26]  Jürgen Schmidhuber,et al.  Long Short-Term Memory , 1997, Neural Computation.

[27]  Kuldip K. Paliwal,et al.  Bidirectional recurrent neural networks , 1997, IEEE Trans. Signal Process..

[28]  Elmar Eisemann,et al.  DeepEyes: Progressive Visual Analytics for Designing Deep Neural Networks , 2018, IEEE Transactions on Visualization and Computer Graphics.

[29]  Mihaela Oprea,et al.  A neural network based model for PM2.5 air pollutant forecasting , 2016, 2016 20th International Conference on System Theory, Control and Computing (ICSTCC).

[30]  Jürgen Schmidhuber,et al.  Applying LSTM to Time Series Predictable through Time-Window Approaches , 2000, ICANN.

[31]  Yoshua Bengio,et al.  Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation , 2014, EMNLP.