537
edits
Theses and Projects: Difference between revisions
Jump to navigation
Jump to search
no edit summary
No edit summary |
No edit summary |
||
| Line 11: | Line 11: | ||
* (M) Master thesis | * (M) Master thesis | ||
* (P) Student project | * (P) Student project | ||
#### 3D natural hazard simulator | |||
The aim of the project is to simulate representative natural hazards for hazard response, such as flooding and forest fire. A natural hazard response simulator will be implemented for both visualization and performance validation. For example, we can visualize the flooding of 2021 in Germany, and then validate the performance of drone deployment in hazard sensing and emergency communication. Here, we introduce some related works in virtual 3D scene which may help you to understand this project, e.g., Agents Toolkit (ML-Agents) [1], DisasterSim [2] and Airsim [3]. | |||
[1] Unity Technologies.Unity ML-Agents Toolkit. Jan 26, 2021.URL:https://github.com/Unity-Technologies/ml-agents. (accessed: 21.11.2021) | |||
[2] Wang, H., Liu, C. H., Dai, Z., Tang, J., & Wang, G. (2021, August). Energy-efficient 3D vehicular crowdsourcing for disaster response by distributed deep reinforcement learning. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (pp. 3679-3687). | |||
[3] S. Shah, D. Dey, C. Lovett, and A. Kapoor. “Airsim: High-fidelity visual and physical simulation forautonomous vehicles”. In:Field and Service Robotics. Springer. 2018, pp. 621–635. | |||
#### Video analytics with deep reinforcement learning | |||
The proliferation of video analytics is facilitated by the advances of deep learning and the low prices of high-resolution network-connected cameras. However, the accuracy improvement from deep learning is at the high computational cost. Although the state-of-the-art smart cameras can support deep learning method, the deployed surveillance and traffic camera paint a much bleaker resource picture. For example, DNNCam that ships with a high-end embedded NVIDIA TX2 GPU costs more than $2000 while the price of deployed traffic cameras today ranges $40-$200; these cameras typically loaded with a single-core CPU only provide very scarce compute resource. Because of this huge gap, typical video analytics applications, e.g., traffic cameras stream live video to remote server for further analysis. | |||
As a result, a natural question occurs: which video streaming configuration also server decoding configuration should we select to guarantee high analysis accuracy as well as not incur network congestion? To answer this question, we attempt to explore the performance of deep reinforcement learning under this scenario. | |||
#### Convolutional neural networks and transfer learning for change estimation with satellite images | |||
Satellite images are a popular input to estimate wealth measures (e.g. income or consumption) on the spatial scale, so to determine which locations are richer or poorer than others within a certain time interval. However, the use of these images for estimation of the changes in these measures over time for given locations is investigated only insufficiently. This project/thesis topic intends to address this problem by applying Convolutional Neural Networks (CNNs) with Transfer Learning on a small data set of images from villages in Thailand and Vietnam. Among other things, this topic contains experimental comparisons of different approaches and CNNs in both Regression and Classification. | |||
#### Socioecomonic analysis based on spatiotemporal and linguistic analysis on microblogging data | |||
Identifying the socioeconomic status (SES) of users in social media like Twitter or Weibo is useful e.g., for digitized advertisements and social policies. This study aims to collect profiles of Twitter users on selected topics such as culture or foreign language learning, extract the temporal, spatial and linguistic features, and compare different classification algorithms (e.g., decision tree, random forest, na\"{i}ve Bayes, deep learning, and Gaussian processes classifier) to predict the socioeconomic status. | |||
[1] Ren Y, Xia T, Li Y, Chen X. Predicting socio-economic levels of urban regions via offline and online indicators. PLoS One. 2019;14(7):e0219058. Published 2019 Jul 10. doi:10.1371/journal.pone.0219058 | |||
[2] Pappalardo L, Pedreschi D, Smoreda Z, Giannotti F. Using Big Data to study the link between human mobility and socio-economic development. In: IEEE International Conference on Big Data 2015. doi:10.1109/BigData.2015.7363835 | |||
[3] Vasileios Lampos, Nikolaos Aletras, Jens K. Geyti, Bin Zou and Ingemar J. Cox (2016). Inferring the Socioeconomic Status of Social Media Users based on Behaviour and Language. Proceedings of the 38th European Conference on Information Retrieval (ECIR '16), pp. 689-695. doi:10.1007/978-3-319-30671-1_54 | |||
=== New video/image encoding for DNN applications === | === New video/image encoding for DNN applications === | ||