No abstract

To reduce the increasingly congestion in cities, it is essential for intelligent transportation system (ITS) to accurately forecast the short-term traffic flow to identify the potential congestion sites. In recent years, the emerging deep learning method has been introduced to design traffic flow predictors, such as recurrent neural network (RNN) and long short-term memory (LSTM), which has demonstrated its promising results. In this paper, different from existing work, we study the temporal convolutional network (TCN) and propose a deep learning framework based on TCN model for short-term city-wide traffic forecast to accurately capture the temporal and spatial evolution of traffic flow. Moreover, we design the model with the Taguchi method to develop an optimized structure of the TCN model, which not only reduces the number of experiments, but also yields high accuracy of forecasting results. With the real-world traffic flow data collected from highways in Birmingham City of U.K., we compare our model with four deep learning based models including LSTM models, GRU models, SAE models, DeepTrend and CNN-LSTM models in terms of the mean absolute error (MAE) and mean relative error (MRE) regarding the actual flow data. The experimental results demonstrate that our framework achieves the state-of-art performance with superior accuracy in short-term traffic flow forecasting. INDEX TERMS Deep learning, temporal convolutional networks, short-term forecasting.

No abstract

Generating stylized captions for images is a challenging task since it requires not only describing the content of the image accurately but also expressing the desired linguistic style appropriately. In this paper, we propose MemCap, a novel stylized image captioning method that explicitly encodes the knowledge about linguistic styles with memory mechanism. Rather than relying heavily on a language model to capture style factors in existing methods, our method resorts to memorizing stylized elements learned from training corpus. Particularly, we design a memory module that comprises a set of embedding vectors for encoding style-related phrases in training corpus. To acquire the style-related phrases, we develop a sentence decomposing algorithm that splits a stylized sentence into a style-related part that reflects the linguistic style and a content-related part that contains the visual content. When generating captions, our MemCap first extracts content-relevant style knowledge from the memory module via an attention mechanism and then incorporates the extracted knowledge into a language model. Extensive experiments on two stylized image captioning datasets (SentiCap and FlickrStyle10K) demonstrate the effectiveness of our method.

No abstract