三菱电机

Neural audio codecs (NACs), which use neural networks to generate compact audio representations, have garnered interest for their applicability to many downstream tasks, especially quantized codecs due to their compatibility with large language models. However, unlike text, speech conveys not only linguistic content but also rich paralinguistic features. Encoding these elements in an entangled fashion may be suboptimal, as it limits flexibility. For instance, voice conversion (VC) aims to convert speaker characteristics while preserving the original linguistic content, which requires a disentangled representation. Inspired by VC methods utilizing k-means quantization with self-supervised features to disentangle phonetic information, we develop a discrete NAC capable of structured disentanglement. Experimental evaluations show that our approach achieves reconstruction performance on par with conventional NACs that do not explicitly perform disentanglement, while also matching the effectiveness of conventional VC techniques.

Recent advances in Large Language Models (LLMs) have improved multi-step reasoning. Most approaches rely on Chain-of-Thought (CoT) rationales. Previous studies have shown that LLMs often generate logically inconsistent reasoning steps even when their final answers are correct. These inconsistencies reduce the reliability of step-level reasoning. We propose GeoSteer, a manifold-based framework that improves the quality of intermediate reasoning. The method consists of: (1) constructing a CoT dataset with segment-level scores, (2) training a Variational Autoencoder (VAE) model and a quality estimation model to learn a low-dimensional manifold of high-quality CoT trajectories, and (3) steering hidden states of target LLMs toward higher-quality regions in the latent space. This update in a latent space behaves like a natural-gradient adjustment in the original hidden-state space. It ensures geometrically coherent steering. We evaluate GeoSteer on the GSM8k dataset using the Qwen3 series. We measure via answer accuracy and overall reasoning performance. GeoSteer improved the exact match accuracy by up to 2.6 points. It also enhanced the pairwise win rate by 5.3 points. These results indicate that GeoSteer provides an effective and controllable mechanism for improving the quality of intermediate reasoning in LLMs.

In the safety management of complex socio-technical systems, an approach based on Safety-II has gained attention, focusing on how systems maintain everyday operations in the face of various disturbances and unexpected situations. Performance variability, which refers to the workers' adjustment activity in response to these external factors, is essential for analyzing system safety and the workers' knowledge about adaptive behaviors. This study proposes a practical method to identify workers' performance variability by comparing Work-as-Imagined (WAI) with Work-as-Done (WAD) and systematically extracting workers' knowledge for managing specific disturbances. This method builds upon the Functional Resonance Analysis Method (FRAM) and associated semi-quantitative simulation techniques developed within the Safety-II paradigm. The approach involves constructing separate FRAM models for WAI and WAD, followed by comprehensive simulations based on a Qualitative Comparative Analysis (QCA) process where functions within the WAD model are iteratively modified. This process identifies the critical functions essential for the operator's adaptive knowledge, providing a systematic framework to capture and analyze workers' resilient performances. The practical application of the proposed approach was demonstrated through a case study of the etching work in a compound semiconductor manufacturing process. The results showed that combining the specific functions in the WAD model contributes to operational resilience, reflecting workers' practical knowledge emerging from interactions between multiple functions. The proposed approach contributes to revising work procedures, improving worker education, and enhancing system safety management.

Real-world scenes often feature multiple humans interacting with multiple objects in ways that are causal, goal-oriented, or cooperative. Yet existing 3D human-object interaction (HOI) benchmarks consider only a fraction of these complex interactions. To close this gap, we present MMHOI – a large-scale, Multi-human Multi-object Interaction dataset consisting of images from 12 everyday scenarios. MMHOI offers complete 3D shape and pose annotations for every person and object, along with labels for 78 action categories and 14 interaction-specific body parts, providing a comprehensive testbed for next-generation HOI research. Building on MMHOI, we present MMHOI-Net, an end-to-end transformer-based neural network for jointly estimating human-object 3D geometries, their interactions, and associated actions. A key innovation in our framework is a structured dual-patch representation for modeling objects and their interactions, combined with action recognition to enhance the interaction prediction. Experiments on MMHOI and the recently proposed CORE4D datasets demonstrate that our approach achieves state-of-the-art performance in multi-HOI modeling, excelling in both accuracy and reconstruction quality.

In the control of a large structure, hydraulic equipment is often used because it requires a large output. In this study, we focus on a telescope as an example of a large structure, and experimentally verify the basic characteristics of the hydraulic support system for the telescope. Large telescopes are equipped with a hydraulic system that adjusts the deformation of the structure, because the change in ambient temperature causes the expansion of the structure, thereby degrading the observational performance of the telescope. In the authors’ previous report, a simple dynamic transfer function model of this system has been derived. In this report, an experimental apparatus simulating the system is built, and the static and dynamic transient response characteristics are evaluated. The transfer function model is evaluated using the experimental response results. As a result, it is confirmed that the system has sufficient properties to achieve a high observational performance of the telescope and that the proposed transfer function model adequately reproduces the properties.