- HOME
- »
- LABORATORIES
- »
- Department of Computer Science
- »
- MINENO Hiroshi
MINENO Hiroshi
Research outline
Our laboratory is concerned with the general problem of developing ubiquitous sensor/actuator network systems, and then making them intelligent based on machine learning, deep learning and reinforcement learning. Mutual complementary techniques with wired/wireless networking play a key role, especially as systems of interest grow in scale. Our target application includes interactions with distributed intelligent informatics and predictive control as well as computer science.
Research on knowledge based smart greenhouse environmental control system using SW-SVR (Sliding Window-based Support Vector Regression) aims to reproduce prediction control performed by expert farmers’ cultivation without human intervention. Our proposed SW-SVR is a new machine learning algorithm for time series data prediction. The system with 429MHz band wireless communication adopted robust sensor network system maintains high packet arrival rate regardless of growth of plants. Our experimental results show the proposed system dramatically reduced prediction error of nitrogen absorption and amount of needed training data.
Research on mobile data 3-dimensional offloading method (MDOP) and high fidelity emulation environment (HiFEE) aims to disperse uplink/downlink mobile data with less work. The rapid growth of smartphone and the wide development of LTE networks have changed the traffic pattern of mobile applications. We propose the MDOP combining with temporal, spacial and channel based offloading methods. To evaluate the performance, we are developing the HiFEE to solve the tradeoff relationship between physical system and network simulator. Virtual machine with virtual wireless LAN device achieves use of real code from MAC management to application layer. In evaluation, HiFEE can reduce the difference of throughput between physical system and simulator.
Research on distributed data stream processing aims to be a platform for spatiotemporal interpolation with adaptive resolution. A hybrid approach of client-side scripting and server-side scripting is one of the key features to balance the load between the client and server. Moreover, data stream direction control is used for receiving data stream directly from the data source. This platform is suitable for developing an application such as monitoring condition of civil infrastructure and visualizing transition of data stream.
Own web site
Data driven smart greenhouse IoT system using AI
Mobile data 3-dimensinal offloading (MDOP) and high fidelity emulation environment (HiFEE)
Distributed data stream processing platform for spatiotemporal interpolation with adaptive resolution