CoDR: Correlation-based Data Reduction Scheme for Efficient Gathering of Heterogeneous Driving Data.

A variety of deep learning techniques are actively employed for advanced driver assistance systems, which in turn require gathering lots of heterogeneous driving data, such as traffic conditions, driver behavior, vehicle status and location information. However, these different types of driving data easily become more than tens of GB per day, forming a significant hurdle due to the storage and network cost. To address this problem, this paper proposes a novel scheme, called CoDR, which can reduce data volume by considering the correlations among heterogeneous driving data. Among heterogeneous datasets, CoDR first chooses one set as a pivot data. Then, according to the objective of data collection, it identifies data ranges relevant to the objective from the pivot dataset. Finally, it investigates correlations among sets, and reduces data volume by eliminating irrelevant data from not only the pivot set but also other remaining datasets. CoDR gathers four heterogeneous driving datasets: two videos for front view and driver behavior, OBD-II and GPS data. We show that CoDR decreases data volume by up to 91%. We also present diverse analytical results that reveal the correlations among the four datasets, which can be exploited usefully for edge computing to reduce data volume on the spot.

Learning Pose-Aware Models for Pose-Invariant Face Recognition in the Wild.

We propose a method designed to push the frontiers of unconstrained face recognition in the wild with an emphasis on extreme out-of-plane pose variations. Existing methods either expect a single model to learn pose invariance by training on massive amounts of data or else normalize images by aligning faces to a single frontal pose. Contrary to these, our method is designed to explicitly tackle pose variations. Our proposed Pose-Aware Models (PAM) process a face image using several pose-specific, deep convolutional neural networks (CNN). 3D rendering is used to synthesize multiple face poses from input images to both train these models and to provide additional robustness to pose variations at test time. Our paper presents an extensive analysis of the IARPA Janus Benchmark A (IJB-A), evaluating the effects that landmark detection accuracy, CNN layer selection, and pose model selection all have on the performance of the recognition pipeline. It further provides comparative evaluations on IJB-A and the PIPA dataset. These tests show that our approach outperforms existing methods, even surprisingly matching the accuracy of methods that were specifically fine-tuned to the target dataset. Parts of this work previously appeared in [1] and [2].

Don't make cache too complex: A simple probability-based cache management scheme for SSDs.

Solid-state drives (SSDs) have recently become a common storage component in computer systems, and they are fueled by continued bit cost reductions achieved with smaller feature sizes and multiple-level cell technologies. However, as the flash memory stores more bits per cell, the performance and reliability of the flash memory degrade substantially. To solve this problem, a fast non-volatile memory (NVM-)based cache has been employed within SSDs to reduce the long latency required to write data. Absorbing small writes in a fast NVM cache can also reduce the number of flash memory erase operations. To maximize the benefits of an NVM cache, it is important to increase the NVM cache utilization. In this paper, we propose and study ProCache, a simple NVM cache management scheme, that makes cache-entrance decisions based on random probability testing. Our scheme is motivated by the observation that frequently written hot data will eventually enter the cache with a high probability, and that infrequently accessed cold data will not enter the cache easily. Owing to its simplicity, ProCache is easy to implement at a substantially smaller cost than similar previously studied techniques. We evaluate ProCache and conclude that it achieves comparable performance compared to a more complex reference counter-based cache-management scheme.

Effective representation using ICA for face recognition robust to local distortion and partial occlusion.

The performance of face recognition methods using subspace projection is directly related to the characteristics of their basis images, especially in the cases of local distortion or partial occlusion. In order for a subspace projection method to be robust to local distortion and partial occlusion, the basis images generated by the method should exhibit a part-based local representation. We propose an effective part-based local representation method named locally salient ICA (LS-ICA) method for face recognition that is robust to local distortion and partial occlusion. The LS-ICA method only employs locally salient information from important facial parts in order to maximize the benefit of applying the idea of "recognition by parts." It creates part-based local basis images by imposing additional localization constraint in the process of computing ICA architecture I basis images. We have contrasted the LS-ICA method with other part-based representations such as LNMF (Localized Nonnegative Matrix Factorization) and LFA (Local Feature Analysis). Experimental results show that the LS-ICA method performs better than PCA, ICA architecture I, ICA architecture II, LFA, and LNMF methods, especially in the cases of partial occlusions and local distortions.