Classification models based on deep neural networks (DNNs) must be calibrated to measure the reliability of predictions. Some recent calibration methods have employed a probabilistic model on the probability simplex. However, these calibration methods cannot preserve the accuracy of pre-trained models, even those with a high classification accuracy. We propose an accuracy-preserving calibration method using the Concrete distribution as the probabilistic model on the probability simplex. We theoretically prove that a DNN model trained on cross-entropy loss has optimality as the parameter of the Concrete distribution. We also propose an efficient method that synthetically generates samples for training probabilistic models on the probability simplex. We demonstrate that the proposed method can outperform previous methods in accuracy-preserving calibration tasks using benchmarks.

Feature representation of oleaginous yeast images is essential for non-invasively observing their cultivation status. However, conventional methods are difficult to capture the visual characteristics of oleaginous yeasts at the extremely local spatial regions. To tackle this problem, this paper proposes micro-spatial attention with sparse constraints for self-supervised learning. Specifically, we newly design a micro-spatial attention module to focus on the extremely local spatial regions for feature representation. Unlike the existing spatial attention module, which requires the guide information (e.g., labels and textual queries) to focus on specific regions, our proposed module employs a sparse constraint. By combining the sparse constraint with self-supervised learning, we can automatically focus on essential regions, which appear in extremely local regions. This improves the feature representation power for oleaginous yeast images with micro-spatial contexts. Experiments on real Lipomyces starkeyi datasets, a specific type of oleaginous yeasts, confirm the superiority of our method. Moreover, our method realized clustering of oleaginous yeast images, enabling us to non-invasively observe the cultivation trends over time.

Pre-trained language models (PLMs) such as BERT and RoBERTa have dramatically improved the performance of various natural language processing tasks. Although these models are trained on large amounts of raw text, they have no explicit grounding in real-world entities. Knowledge graphs (KGs) are manually annotated with factual knowledge and store the relations between nodes corresponding to entities as labeled edges. This paper proposes a mechanism called KG-attention, which integrates the structure of a KG into recent PLM architectures. Unlike the existing PLM+KG integration methods, KG-attention generalizes the embeddings of neighboring entities using the relation embeddings; accordingly, it can handle relations between unconnected entities in the KG. Experimental results demonstrated that our method achieved significant improvements in a relation classification task, an entity typing task, and several language comprehension tasks.

Redirected walking has been proposed as a means of making a narrow space feel wide in a virtual space. We examined the effect of visuo-haptics on the detection threshold of rotation gain when participants walked around a wall. We found that the threshold was affected by visuo-haptics only when they walked around the outside of the wall but was not affected when they walked around the inside of the wall.

Smoke is an important sign for early fire detection. Image based detection methods are more useful than other methods which use some special sensor devices. When treating image information of smoke, it is important to consider characteristics of smoke. In this study, we consider that the image information of smoke is a self-affine fractal. We focus on the nature of smoke and present a new smoke detection method based on the fractal property of smoke. We use the Hurst exponent H, which is one of the widely known exponent of fractals. We calculate H of smoke from a relation between H and the wavelet transform of the image. So we detect smoke areas in images with H through the wavelet transform. Moreover, to obtain the accurate detection result, we use the time-accumulation technique to smoke detection results of each image. In experiments, we show the effectiveness of our method with the fractal property of smoke.

This paper presents a novel prediction based digital control dc-dc converter. In this method, addition to the P-I-D control as the feedback control, the prediction based control is used as the feedfoward control. In the feedfoward control, the neural network based method is adopted. This works to improve the transient response very effectively when the load is changed quickly. As a result, the undershoot of the output voltage and the overshoot of the reactor current are suppressed effectively as compared with the conventional one in the step change of load resistance. It is confirmed that the prediction based control technique is useful to realize the high performance digital control method for the dc-dc converter.

The purpose of this paper is to present a new digital control method of the forward type multiple-output DC DC converter with both a PID feedback and a new feed forward control. In this converter, two novel control methods are proposed. The first new control method is a model method and the second new method is a neural network predictor. The dynamic characteristics of digital control DC DC converter are improved as compared with the conventional one. Especially, the digital control DC DC converter with method of the neural network can be realized excellent dynamic characteristics. As a result, the undershoot of the output voltage and the overshoot of reactor current are improved to 45% and 26%, respectively.

The paper describes a stabilization control of two wheels driven wheelchair based on pitch angle disturbance observer (PADO). PADO makes it possible to stabilize the wheelchair motion and remove casters. This brings a sophisticated mobility of wheelchair because the casters are obstacle to realize step passage motion and so on. The proposed approach based on PADO is robust against disturbance of pitch angle direction and the more functional wheelchairs is expected in the developed system. The validity of the proposed method is confirmed by simulation and experiment.