Akira Ishii

Hakaze Cho, Yoshihiro Sakai, Mariko Kato et al.

In-Context Learning (ICL) typically utilizes classification criteria from output probabilities of manually selected label tokens. However, we argue that such token-based classification criteria lead to suboptimal decision boundaries, despite delicate calibrations through translation and constrained rotation applied. To address this problem, we propose Hidden Calibration, which renounces token probabilities and uses the nearest centroid classifier on the LM's last hidden states. In detail, we assign the label of the nearest centroid previously estimated from a calibration set to the test sample as the predicted label. Our experiments on 6 models and 10 classification datasets indicate that Hidden Calibration consistently outperforms current token-based baselines by about 20%~50%, achieving a strong state-of-the-art in ICL. Our further analysis demonstrates that Hidden Calibration finds better classification criteria with less inter-class overlap, and LMs provide linearly separable intra-class clusters with the help of demonstrations, which supports Hidden Calibration and gives new insights into the principle of ICL. Our official code implementation can be found at https://github.com/hc495/Hidden_Calibration.

Takato Yasuno, Akira Ishii, Masazumi Amakata

Recently, flood damage has become a social problem owing to unexperienced weather conditions arising from climate change. An immediate response to heavy rain is important for the mitigation of economic losses and also for rapid recovery. Spatiotemporal precipitation forecasts may enhance the accuracy of dam inflow prediction, more than 6 hours forward for flood damage mitigation. However, the ordinary ConvLSTM has the limitation of predictable range more than 3-timesteps in real-world precipitation forecasting owing to the irreducible bias between target prediction and ground-truth value. This paper proposes a rain-code approach for spatiotemporal precipitation code-to-code forecasting. We propose a novel rainy feature that represents a temporal rainy process using multi-frame fusion for the timestep reduction. We perform rain-code studies with various term ranges based on the standard ConvLSTM. We applied to a dam region within the Japanese rainy term hourly precipitation data, under 2006 to 2019 approximately 127 thousands hours, every year from May to October. We apply the radar analysis hourly data on the central broader region with an area of 136 x 148 km2 . Finally we have provided sensitivity studies between the rain-code size and hourly accuracy within the several forecasting range.

Takato Yasuno, Akira Ishii, Junichiro Fujii et al.

In order to monitor the state of large-scale infrastructures, image acquisition by autonomous flight drones is efficient for stable angle and high-quality images. Supervised learning requires a large data set consisting of images and annotation labels. It takes a long time to accumulate images, including identifying the damaged regions of interest (ROIs). In recent years, unsupervised deep learning approaches such as generative adversarial networks (GANs) for anomaly detection algorithms have progressed. When a damaged image is a generator input, it tends to reverse from the damaged state to the healthy state generated image. Using the distance of distribution between the real damaged image and the generated reverse aging healthy state fake image, it is possible to detect the concrete damage automatically from unsupervised learning. This paper proposes an anomaly detection method using unpaired image-to-image translation mapping from damaged images to reverse aging fakes that approximates healthy conditions. We apply our method to field studies, and we examine the usefulness of our method for health monitoring of concrete damage.