【発表】後期博士課程 礒川君が研究内容を国際会議Empowering the Lifestyle and Well-being of Eastern Asian Elders with intelligent IoTで発表しました – ‘ Landmark Point Detection for Facial Expression Recognition in Dogs’
【発表】修士1年生沖原君が研究内容を国際会議Empowering the Lifestyle and Well-being of Eastern Asian Elders with intelligent IoTで発表しました – ‘ Material Recognition by Passive Sensing of Vibration of an Object Placed on a Surface Using a Portable Accelerometer’
【採択】学部3年生 泉川君の研究が国際会議ACM HotMobile 2023 poster sessionにacceptされました – ‘ Audio-based Eating Stage Recognition through CNN Model Trained on ASMR Eating Sounds’
Abstract:
A balanced diet and an appropriate calorie intake are the keys to both preventing and treating type II diabetes. Meanwhile, widespread techniques such as manual food logs and food image captures have been posing burdens on those with diabetes and have made diet monitoring difficult to become part of one’s routine. To develop an earable device that monitors a volume of food intake automatically, an-audio based detection of eating instances is necessary. The present research therefore attempted to classify an eating sound, collected from YouTube eating ASMR, into one of the following labels: chew, swallow, or non-eating A CNN machine learning model using sound features as input achieved an accuracy of 81%.
We propose a 3D CO2 concentration analyzing system to monitor the ventilation of the closed space. This system uses a 3D scan to lower the labor of creating a 3D model of the space. In this research, multiple CO2 sensors are used, and consider the model to map point CO2 concentration data to the spatial CO2 distribution. We experimented to check the accuracy of the model in the real bus space.
【採択】修士1年生浜中君の研究が国際会議ACM HotMobile 2023 poster sessionにacceptされました – ‘ Estimation of user personality traits on the Web Using Multi-Task Learning’
Abstract:
We focus on user’s personality traits and propose a method to estimate user’s personality traits from footprint logs collected from news platform services and search queries, with the aim of constructing a news recommendation system using the personality traits. We constructed supervised machine learning models for 8871 users, using the features calculated from news browsing logs as explanatory variables and personality traits (Big Five) collected via a questionnaire using crowdsourcing as objective variables. Five estimators were constructed for each Big Five traits, and the results of comparative evaluation using multiple algorithms showed that the accuracy of AUC 0.634 was obtained for the agreeableness trait.