This paper is a description of human mobility prediction system which was submitted for 14th SIGSPATIAL Cup competition (GISCUP 2025). The system uses common Bidirectional Encoder Representations from Transformers (BERT) architecture. However, there are two distinctive features. The first is that it performs data augmentation by rotating the human mobility trajectory in provided training data. The second is that the system applies ensemble from each epoch during the model training. According to the results of evaluation experiments using held out data, the system improves Geo-BLEU by 17.9% on average, with up to 21.4% improvement for individual cities compared to the conventional BERT-based method.

Understanding fine-grained urban population distribution based on GPS location data is important for urban applications such as traffic management and new store openings for retailers. However, GPS-based population distribution relies heavily on the number of users who agree to provide GPS logs. With only a limited number of users, the fine-grained population distribution becomes sparse and must be aggregated as coarse-grained. In this paper, we present the challenge of developing a model to estimate fine-grained population distributions from coarse-grained population distributions and propose a model capable of incorporating extensive auxiliary information using a CNN-based image super-resolution approach. Our experiments with real data reveal two key findings: (i) traditional regression models tend to estimate similar populations for adjacent grids, which is often overlooked by existing metrics, and (ii) CNN-based image super-resolution models reproduce population distribution features of adjacent grids having different population volumes, although they sometimes provide simplistic estimates depending on auxiliary information. Based on these findings, we present our vision for developing a promising model and improving the evaluation metrics tailored to this challenge.

Previous studies have shown that the personality composition of a group significantly affects learners’ satisfaction during collaborative learning. However, while these studies investigated a group as a whole by focusing on group statistics, such as the mean and standard deviation of the members’ personalities, they paid little attention to the personality differences of individual pairs within the group, albeit the group contains many pairwise interactions. In this paper, we studied whether and how pairwise personality differences between a learner and groupmates affect the learner’s satisfaction. Examining data collected from an employee training program during which learners had reflective group discussions, we confirmed that pairwise personality differences significantly affect a learner’s level of satisfaction in the program. Specifically, satisfaction is affected by (1) the average of the personality differences between the learner and each individual groupmate, which reflects the degree to which the learner is different from the groupmates on average, and (2) the personality difference from the groupmate who has the most different/similar personality from/to the learner.

In this paper, we propose an approach to predict customers’ personalities leveraging two modalities of customers’ data: (a) service usages logs of online services and (b) visual data collected in physical stores by surveillance cameras (e.g., gait and whereabouts). A number of companies provide services via online and offline nowadays, thus need to serve two different kinds of customers: “online customers,” who use services completely online and have only (a); and “offline customers,” who use only physical stores and have only (b). To improve personality prediction accuracy for these customers, our approach generates pseudo features of a non-existent modality from the other modality that the customers actually have (i.e., feature projection; e.g., generate pseudo visual data of the online customers from their real online service logs), and uses both pseudo and real features to predict their personalities. The evaluation using real-world data of a mobile carrier’s customers showed that our approach predicted personality more accurately than an ordinary unimodal approach for both online and offline customers. We also examined importance of the feature projection and compared two different projection methods.

In this paper, we propose an online 3D person tracking method and an attribute estimation method with pedestrian dead reckoning (PDR). For person tracking, we employ a structured prediction approach, which extends the Struck algorithm. Although the main stream of visual object tracking, including Struck, utilizes only 2D information in image coordinates, it is difficult to track object correctly because of changes in the scale and angle of the target. In contrast, our classifier adaptively learns structural relationship in world coordinates and in image coordinates using Structured SVM. Furthermore, we combine visual tracking results and sensor trajectories based on PDR. Our method estimates a person attribute whether insider like a sales staff, or outsider like a customer. According to experimental results, the proposed method outperforms the existing methods regarding the quality of localization. In addition, experimental results show that our method can estimate the attribute at a ratio of 0.84.

CDMA2000 lxEV-DO is a code division multiplex cellular system and requires that the terminals search the PN code sequences of the base station (BS) during the initial acquisition stage. The exclusive hardware is frequently used for initial acquisition due to its heavy computational complexity. However, in software defined radio (SDR) terminals, initial acquisition has to be processed by software using the digital signal processor (DSP), which has reasonable performance, and must be finished within a practical time. This paper presents an initial acquisition scheme suitable for SDR-based CDMA2000 lxEV-DO terminals. The implementation onto our SDR platform and the field test results of the initial acquisition time are also described.

The traditional software defined radio (SDR) implementations use either a high-performance Field Programmable Gate Array (FPGA) or multiple high-speed Digital Signal Processors (DSPs) to execute digital signal processing, consuming a large amount of electrical power. To realize an SDR mobile handset, it is necessary to reconcile the two contradictory requirements; high performance able to execute digital signal processing in real time, and low power consumption to allow battery operation. This paper presents the software radio implementation of a CDMA2000 1xEV-DO mobile handset on a single low-power- consumption DSP chip. By overcoming difficulties in multicore and multithread architecture, in parallel processing algorithm design, and in processing time restrictions, we have confirmed that our implementation has the ability to execute real-time signal processing up to 2.4 Mbps with a maximum data rate of lxEV-DO, and to communicate with a base station sending and receiving IP packets.