Large language models (LLMs) have achieved notable success across various tasks but are hindered by their large size and high computational demands. Post-training pruning (PTP) offers a promising solution by reducing model size through parameter removal while preserving performance. However, current PTP methods perform optimally only within specific sparsity ranges. This paper presents two key findings: (1) Layerwise uniform sparsity is effective at low sparsity, while non-uniform sparsity excels at high levels; (2) Relative importance-based pruning works best at low sparsity, whereas Hessian-based weight reconstruction is superior at high sparsity. We design and conduct experiments to validate these findings. Based on these insights, we introduce OptiPrune, a robust pruning method effective across all sparsity levels. OptiPrune adapts non-uniform sparsity with adaptive deviation and employs a threshold to select the optimal pruning strategy. Empirical results across diverse datasets, architectures, and languages validate its performance and robustness. These findings provide valuable directions for future LLM pruning research. Our code and data are publicly available.

With the aim to generate beneficial reports to improve the content of walking tourism, this paper discusses methods to visualize tourist interest in geographic features in a city by automatically collecting their feedback through mobile applications. Geo-density maps that use raw trajectories generally have difficulty in determining whether the dense areas presented on them are more appealing than the other areas in a city, since environmental, user-behavioral, and technical factors present various degrees of density regardless of tourist activity. Additionally, the complicated point patterns in a trajectory underestimate polyline-shaped features, such as footpaths and streets, which are some of the most important components of walking tourism, and their density maps prevent the inference of hot streets. The proposed framework monitors user behavior, e.g., entering/leaving indoor spaces and walking/stopping, using various mobile sensors to generate semi-ready trajectory data for thematic heatmaps of city-scale tourist analysis by (1) equalizing the inherent density bias in a trajectory, (2) mitigating the errors of GPS recording, (3) adding attributes that describe concrete user events during tourism. Furthermore, our client system demonstrates hot street heatmaps, user-generated content (UGC)-oriented hot spot heatmaps, and indoor-oriented hot spot heatmaps drawn from semi-ready data. The evaluation experiment conducted in this paper illustrates that the proposed data were able to represent the actual walking trajectory more accurately than the raw GPS trajectory and existing line simplification approaches.

Although application examples of multilevel optimization have already been discussed since the 1990s, the development of solution methods was almost limited to bilevel cases due to the difficulty of the problem. In recent years, in machine learning, Franceschi et al. have proposed a method for solving bilevel optimization problems by replacing their lower-level problems with the $T$ steepest descent update equations with some prechosen iteration number $T$. In this paper, we have developed a gradient-based algorithm for multilevel optimization with $n$ levels based on their idea and proved that our reformulation asymptotically converges to the original multilevel problem. As far as we know, this is one of the first algorithms with some theoretical guarantee for multilevel optimization. Numerical experiments show that a trilevel hyperparameter learning model considering data poisoning produces more stable prediction results than an existing bilevel hyperparameter learning model in noisy data settings.

In this study, we suggest a method to coordinate turn-taking and talking in multi-party conversations by the gaze of a robot that participates on the side. Also, we use the experimental paradigm named “Cooperative Turn-taking Game in Non-verbal Situation”, which is a simplified multi-party conversation environment. We investigated whether designing eye-gazes for such robots can coordinate turn-taking and talking in multi-party conversations, and we found the robot's gaze could coordinate turn-taking and talking in multi-party conversations. Our study is expected to effectively encourage desirable talking in such multi-party conversations as collaborative learning scenes.

Computer modeling of a large-scale scene such as a city becomes an important topic for computer vision and computer graphics research areas etc. Image-based rendering (IBR) is an effective method for expressing realistic scene, and can construct any arbitrary viewpoint by using the captured real images. However, the large size of the image database in IBR causes serious problems in actual applications, leading to the use of compression techniques. We propose a compression technique based on eigen space combined with a block matching technique to get better result. We also propose a technique to restore the compressed data on Graphic Processing Unit (GPU), allowing us to perform high-speed rendering without raising the load on the CPU.

This paper presents a method for deciding the timing of turn-taking in spoken dialogue systems. This method uses a decision tree learned from the corpus of dialogues between human users and systems in which desirable turn-taking behaviors are annotated by hand. It utilizes a variety of attributes, such as recognition and understanding results and prosodic information. Unlike most of the existing systems it enables spoken dialogue systems to decide the timing of turn-taking based on not only pauses but also other features, so that users can speak to the system even if they put pauses in the middle of their utterances. The result of a preliminary experiment shows that the learned decision tree outperforms the baseline strategy, which takes turn at every user pauses. 1.

For part I see ibid., p.1-10. Dataflow diagrams and entity life histories are widely-used tools in information systems methodologies. The tools are far more effective than natural language, though their meanings have not yet been clarified. This paper shows how the model developed in part I enables formal characterization of information systems methodologies. That is, it elaborates the meaning of DFDs and ELHs by providing a concrete example throughout. The characterization of DFDs establishes an explicit correspondence between the static structure of a business transaction system and a DFD. This correspondence illustrates the necessary components in designing the file structure of a business system. The ELH of a business system is described as the set of sequences of the system's transactions. Black hole and franchise business systems can be characterized by sequences, while the ELH of a general business system partially defines its dynamic structure.

A variety of information systems methodologies for information systems analysis have been proposed. Though each methodology has its own effective concepts and tools, there still does not seem to exist a rigorous model of a business system. In this paper a general model of business systems is proposed. The model describes the whole mechanism of typical routine processing of business tasks with slips and business papers. Furthermore it can be used to examine the ability and limitation of popular tools. The model proposed is called a business transaction system. It consists of both static and dynamic structures. The former depicts the interconnection of the transactions and file system in a business system. The dynamic structure is constructed as a state space representation by introducing a state space, and then the resultant dynamic system is a discrete event system. The state space consists of the file system and schedule of transactions processing. The model provides answers to some questions about the nature of information systems methodologies.