Traffic accidents, mostly caused by human errors, pose a significant challenge in automotive safety. This paper introduces TUN-DAS, a novel Time-series analysis and Unsupervised learning-based Driving behavior Assessment System, designed to systematically evaluate drivers’ behavior and identify potential risks. Unlike conventional in-cabin driver monitoring systems, which generally rely on rule-based or supervised methods, TUN-DAS utilizes a unique time-series clustering approach. This method combines a k-means algorithm with Dynamic Time Warping (DTW) and DTW Barycenter Averaging (DBA), enabling a more nuanced and comprehensive assessment of driving behavior. The modification of conventional DTW in our system allows for effective computation of temporal differences between driving data. TUN-DAS stands out in its ability to detect both sporadic human errors and consistent unsafe driving patterns by analyzing anomalies within individual drivers and across a broader driver dataset. Our validation with bus driver data shows TUN-DAS’s effectiveness in driving behavior assessment, significantly achieving better accuracy than traditional rule-based evaluation methods and allowing for quantitative and temporal analysis of a series of driving behavior. This system holds promise for enhancing road safety by facilitating the early identification of high-risk drivers, thereby contributing to the development of safer driving environments.

With thousands of academic articles shared on a daily basis, it has become increasingly difficult to keep up with the latest scientific findings.To overcome this problem, we introduce a new task of disentangled paper summarization, which seeks to generate separate summaries for the paper contributions and the context of the work, making it easier to identify the key findings shared in articles.For this purpose, we extend the S2ORC corpus of academic articles, which spans a diverse set of domains ranging from economics to psychology, by adding disentangled "contribution" and "context" reference labels.Together with the dataset, we introduce and analyze three baseline approaches: 1) a unified model controlled by input code prefixes, 2) a model with separate generation heads specialized in generating the disentangled outputs, and 3) a training strategy that guides the model using additional supervision coming from inbound and outbound citations.We also propose a comprehensive automatic evaluation protocol which reports the relevance, novelty, and disentanglement of generated outputs.Through a human study involving expert annotators, we show that in 79%, of cases our new task is considered more helpful than traditional scientific paper summarization.

These days, value structure and social structure are changing with the background of immigration, globalization, and diversification. In many countries, such as Japan, a foreign workforce is introduced due to ageing citizens and labour shortages. Such diversification would be a great opportunity to create innovation. Innovation is often achieved when ideas from different perspectives synergize with each other. In this paper, we focus on the job hunting scene and suggest recruiting diverse and cooperative teams as one form of recruitment. Conventional job-hunting services and Social networking services (SNSs) are limited to matching labour supply and demand or attracting people with similar values. To help build and recruit diverse and cooperative teams, we propose a system to build a platform to connect people with different backgrounds, skills, and values and promote their cooperation. With the system, job hunters could find a team where members can leverage each other’s strengths and compensate for each other’s weaknesses. Furthermore, companies could effectively evaluate the diversity and cooperativeness of teams for hiring decisions. The evaluation scenario demonstrated that the proposed system could increase both job hunter and recruiter satisfaction levels and social impact.

It has been shown that machine translation models usually generate poor translations for named entities that are infrequent in the training corpus.Earlier named entity translation methods mainly focus on phonetic transliteration, which ignores the sentence context for translation and is limited in domain and language coverage.To address this limitation, we propose DEEP, a DEnoising Entity Pretraining method that leverages large amounts of monolingual data and a knowledge base to improve named entity translation accuracy within sentences.Besides, we investigate a multi-task learning strategy that finetunes a pre-trained neural machine translation model on both entity-augmented monolingual data and parallel data to further improve entity translation.Experimental results on three language pairs demonstrate that DEEP results in significant improvements over strong denoising autoencoding baselines, with a gain of up to 1.3 BLEU and up to 9.2 entity accuracy points for English-Russian translation. 1

Zi-Yi Dou, Pengfei Liu, Hiroaki Hayashi, Zhengbao Jiang, Graham Neubig. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Abstract Aspect-based summarization is the task of generating focused summaries based on specific points of interest. Such summaries aid efficient analysis of text, such as quickly understanding reviews or opinions from different angles. However, due to large differences in the type of aspects for different domains (e.g., sentiment, product features), the development of previous models has tended to be domain-specific. In this paper, we propose WikiAsp,1 a large-scale dataset for multi-domain aspect- based summarization that attempts to spur research in the direction of open-domain aspect-based summarization. Specifically, we build the dataset using Wikipedia articles from 20 different domains, using the section titles and boundaries of each article as a proxy for aspect annotation. We propose several straightforward baseline models for this task and conduct experiments on the dataset. Results highlight key challenges that existing summarization models face in this setting, such as proper pronoun handling of quoted sources and consistent explanation of time-sensitive events.

In this paper, we propose Latent Relation Language Models (LRLMs), a class of language models that parameterizes the joint distribution over the words in a document and the entities that occur therein via knowledge graph relations. This model has a number of attractive properties: it not only improves language modeling performance, but is also able to annotate the posterior probability of entity spans for a given text through relations. Experiments demonstrate empirical improvements over both word-based language models and a previous approach that incorporates knowledge graph information. Qualitative analysis further demonstrates the proposed model's ability to learn to predict appropriate relations in context. 1

Distracted driving is one of the most common causes of traffic accidents around the world. Recognizing the driver's gaze direction during a maneuver could be an essential step for avoiding the matter mentioned above. Thus, we propose a gaze zone classification system that serves as a base of supporting systems for driver's situation awareness. However, the challenge is to estimate the driver's gaze inside not ideal scenarios, specifically in this work, scenarios where may occur self-occlusions or non-aligned head and eyes direction of the driver. Firstly, towards solving miss classifications during self-occlusions scenarios, we designed a novel protocol where a 3D full facial geometry reconstruction of the driver from a single 2D image is made using the state-of-the-art method PRNet. To solve the miss classification when the driver's head and eyes direction are not aligned, eyes and head information are extracted. After this, based on a mix of different data pre-processing and deep learning methods, we achieved a robust classifier in situations where self-occlusions or non-aligned head and eyes direction of the driver occur. Our results from the experiments explicitly measure and show that the proposed method can make an accurate classification for the two before-mentioned problems. Moreover, we demonstrate that our model generalizes new drivers while being a portable and extensible system, making it easy-adaptable for various automobiles.

Highly-automated vehicles operating in level 3 issue a takeover request (TOR) to transfer the control authority from the automated driving (AD) system to a human driver when they encounter system limitations. In such `unscheduled' situations, the driver is required to immediately re-engage in the driving task both physically and cognitively, and perform suitable action, e.g. lane change. Thus, evaluating driver engagement by the AD system would lead to safe takeover. Physical engagement is easily estimated but there are few studies on evaluating cognitive engagement. In this study, we thus developed a driver situational awareness estimation system based on glance information. We first defined seven standard glance areas and driver glance classification model using a convolutional neural network. We then obtained a large amount of glance data when both safe and dangerous takeover situations (lane change) by using a driving simulator, and we derived the standard glance model including the glance area and time, in order to estimate whether driver gained enough cognitive re-engagement in real-time. To evaluate the effectiveness of the proposed model, we created a situational awareness assist system to visually indicate regions with insufficient glance. As a result, we found that the assist system drastically improved driving performance and reduced the number of accidents during takeover.

The present study investigates a Medium Access Control (MAC) protocol for reliable inter-vehicle communications (IVC) to support safe driving with the goal of reducing road traffic accidents. A number of studies have evaluated the performance of the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol. However, the communication quality provided by the CSMA/CA protocol is degraded seriously by the hidden terminal problem in IVC. Therefore, we propose a new MAC protocol, referred to as Periodic BroadcastTiming Reservation Multiple Access (PB-TRMA), which can autonomously control transmission timing and avoid packet collisions by enhancing the Network Allocation Vector (NAV) for periodic broadcast communications. The simulation results show that the proposed protocol can resolve the hidden terminal problem and mitigate data packet collisions. Moreover, the behavior of PB-TRMA is similar to that of Time-Division Multiple Access (TDMA).

We propose a character size optimization technique to enhance the throughput of maskless lithography as well as photomask manufacture. The number of electron beam shots to draw the patterns of circuits is a dominant factor in the manufacture time and the cost for devices. Our technique is capable of drastically reducing them by optimizing the size of characters, which are the patterns to project and are placed on CP masks. Experimental results show that our technique reduced 72.0% of EB shots in the best case, comparing with the ad hoc character sizing