Theses and Projects: Difference between revisions
Haihan.zhang (talk | contribs) |
|||
| (34 intermediate revisions by 12 users not shown) | |||
| Line 4: | Line 4: | ||
[http://www.net.informatik.uni-goettingen.de/?q=news/annual-report-2020-best-wishes-2021 annual report(s)] for our recent activities. | [http://www.net.informatik.uni-goettingen.de/?q=news/annual-report-2020-best-wishes-2021 annual report(s)] for our recent activities. | ||
== Joint Thesis Topics with fortiss == | |||
The following Master's thesis topics are offered in collaboration with Prof. Rute Sofia at fortiss, the Research Institute of the Free State of Bavaria for Software-Intensive Systems, in Munich. | |||
=== * '''New!''' Cognitive Network Intelligence for Energy-Efficient and Deterministic Industrial Communication === | |||
Industrial applications increasingly rely on heterogeneous networks combining 5G, Wi-Fi, and Time-Sensitive Networking (TSN). This topic investigates energy-aware traffic engineering to reduce network energy consumption while meeting strict latency, jitter, and reliability requirements. Tasks include developing energy models, designing optimization mechanisms, and evaluating them using the ns-3 simulation framework. We expect you to have programming skills in Python or C/C++ and an interest in network optimization. A background in wireless communications, industrial networking, or network simulation is an advantage. | |||
[https://www.fortiss.org/en/career/masters-thesis-cognitive-network-intelligence-for-energy-efficient-and-deterministic-industrial-communication Further information] | |||
=== * '''New!''' Energy-Efficient IoT-Edge-Cloud Infrastructures for Dynamic AI Services === | |||
Deploying AI inference across heterogeneous edge and cloud resources requires balancing inference quality, latency, and energy consumption. This topic focuses on energy-aware scheduling of small large language models (LLMs), building on the energy monitoring infrastructure in the fortiss Industrial IoT Lab. Tasks include characterizing inference energy consumption and developing scheduling strategies across different hardware configurations and model quantization levels. We expect you to have a background in machine learning, distributed systems, or computer networking, together with programming skills. Familiarity with PyTorch or Hugging Face is an advantage. | |||
[https://www.fortiss.org/en/career/masters-thesis-energy-efficient-iot-edge-cloud-continuum-infrastructures-for-dynamic-ai-service-support Further information] | |||
=== * '''New!''' Decentralised Federated Learning for Semantic Communications === | |||
Semantic communication aims to reduce communication overhead by transmitting information relevant to a specific inference or control task. This topic investigates decentralised federated learning for training and adapting semantic encoder-decoder models across heterogeneous industrial IoT, edge, and cloud environments. Tasks include developing a distributed learning architecture, addressing non-IID data and device heterogeneity, and evaluating task performance and communication overhead through simulation. We expect you to have programming skills and a background in machine learning or distributed systems. Familiarity with federated learning frameworks or ns-3 is an advantage. | |||
[https://www.fortiss.org/en/career/masters-thesis-decentralised-federated-learning-for-semantic-communications-in-iiot-edge-cloud-environments Further information] | |||
=== * '''New!''' Semantic-Aware Routing for Multi-Hop Industrial IoT Networks === | |||
Routing decisions in industrial IoT networks can affect how well transmitted information supports an application’s task. This topic investigates routing mechanisms that account for task requirements and semantic distortion across multiple network hops. Tasks include designing a protocol that uses distortion budgets and contextual knowledge to select forwarding paths, implementing it in ns-3, and evaluating it against conventional routing approaches. We expect you to have a background in computer networking or wireless communications and programming skills in Python and C/C++. Familiarity with network simulation is an advantage. | |||
[https://www.fortiss.org/en/career/masters-thesis-semantic-aware-routing-for-multi-hop-iiot-infrastructures Further information] | |||
For more information about joint supervision, please contact Prof. Xiaoming Fu [fu@cs.uni-goettingen.de]. | |||
== Joint PhD Program with University of Sydney == | == Joint PhD Program with University of Sydney == | ||
| Line 21: | Line 52: | ||
* (P) Student project | * (P) Student project | ||
=== * '''New!''' Using LLM for Sign Language Translation (B/M/P)=== | |||
Sign language is the primary means of communication for the deaf and hard-of-hearing community, yet most people do not understand it. This topic explores the integration of Large Language Models (LLMs) with computer vision to build an advanced sign language translation system - with special focus on overcoming the critical challenge of understanding long, continuous sign language videos. We welcome students passionate about Natural Language Processing (NLP) and Computer Vision (CV) to explore the cutting edge of sign language translation technology. | |||
Please contact Wenfang Wu [wenfang.wu@cs.uni-goettingen.de] | |||
=== * '''New!''' Using LLM for Sentiment Knowledge Graph Construction (B/M/P)=== | |||
Constructing a sentiment knowledge graph using Large Language Models (LLMs) like ChatGPT involves leveraging the model's capabilities to understand and analyze textual data, extract entities and relationships, perform sentiment analysis, and organize the information into a graph structure. We expect you have a background in knowledge graph and programming skills in Python. | |||
Please contact Wenfang Wu [wenfang.wu@cs.uni-goettingen.de] | |||
Please contact | === * '''New!''' Evaluating Generalization in Vision–Language–Action Models (B/M/P) === | ||
A robot that retrieves a cup in a familiar scene may miss it when the camera moves, even though the cup and the instruction stay the same. How much does a vision–language–action policy depend on its usual view of the workspace or the starting pose of its arm? | |||
You will evaluate an existing policy on a simulated pick-and-place task. First, establish how often the robot places the object in the target area under the standard conditions. Then vary the camera viewpoint and the initial arm pose separately, using matched object placements and the same trial budget. Record where each attempt fails during the approach, grasp or placement. You will deliver a reproducible evaluation script, a comparison of success rates and a set of annotated failure examples. | |||
For a student project, choose one of the two changes. For a Bachelor's thesis, evaluate one change systematically and analyze the resulting failure cases. For a Master's thesis, study both and test whether a recovery step helps when they occur together. You should be comfortable programming in Python and have a basic understanding of machine learning. Experience with PyTorch or a robotics simulator is helpful. | |||
Please contact Haihan Zhang [haihan.zhang@cs.uni-goettingen.de] | |||
=== * '''New!''' Cooperative Perception and Resource Optimization in Vehicular Networks (B/M/P)=== | |||
Cooperative perception enables vehicles and roadside infrastructure to share sensing information and better understand their surroundings. This topic explores joint sensing, communication, and computing resource optimization to improve perception performance while reducing latency and resource consumption. We welcome students with a background in wireless networks, edge computing, optimization algorithms, and Python programming. | |||
Please contact [Yuxin Zhang] [yuxin.zhang@stud.uni-goettingen.de]. | |||
=== * 3D natural hazard simulator === | === * 3D natural hazard simulator === | ||
| Line 69: | Line 112: | ||
Please contact Dr. Tingting Yuan [tingting.yuan@cs.uni-goettingen.de]] (B/M/P) | Please contact Dr. Tingting Yuan [tingting.yuan@cs.uni-goettingen.de]] (B/M/P) | ||
=== * | === * '''[Closed]''' Privacy-preserved Video Analytics=== | ||
This project/thesis topic focuses on the protection of privacy in video analytics. | This project/thesis topic focuses on the protection of privacy in video analytics. | ||
| Line 84: | Line 127: | ||
[1] Pecam: privacy-enhanced video streaming and analytics via securely-reversible transformation [https://dl.acm.org/doi/abs/10.1145/3447993.3448618]. | [1] Pecam: privacy-enhanced video streaming and analytics via securely-reversible transformation [https://dl.acm.org/doi/abs/10.1145/3447993.3448618]. | ||
=== * | === * '''[Closed]''' AI for networking adaption === | ||
In this project/theses topic, you will explore how to make AI meets networking requirements (e.g., fluctuating network states). | In this project/theses topic, you will explore how to make AI meets networking requirements (e.g., fluctuating network states). | ||
You will (1) deploy and test Genet[1]; (2)extend the Genet environment to multi-client environment (e.g., ABR); (3) deploy multi-agent algorithms on Genet and valid the performance. | You will (1) deploy and test Genet[1]; (2)extend the Genet environment to multi-client environment (e.g., ABR); (3) deploy multi-agent algorithms on Genet and valid the performance. | ||
| Line 109: | Line 152: | ||
Please contact Dr. Tingting Yuan [tingting.yuan@cs.uni-goettingen.de] (B/M/P) | Please contact Dr. Tingting Yuan [tingting.yuan@cs.uni-goettingen.de] (B/M/P) | ||
=== * ''' | === * '''[Closed]''' 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. | 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. | ||
| Line 118: | Line 161: | ||
Please contact Prof. Xiaoming Fu [fu@cs.uni-goettingen.de](B/M/P) | Please contact Prof. Xiaoming Fu [fu@cs.uni-goettingen.de](B/M/P) | ||
=== [Closed] Super resolution technique for efficient video delivery === | === [Closed] Super resolution technique for efficient video delivery === | ||
| Line 157: | Line 199: | ||
Please contact Dr.Tingting Yuan [tingting.yuan@cs.uni-goettingen.de ] and Weijun Wang [weijun.wang@informatik.uni-goettingen.de](B/M/P) | Please contact Dr.Tingting Yuan [tingting.yuan@cs.uni-goettingen.de ] and Weijun Wang [weijun.wang@informatik.uni-goettingen.de](B/M/P) | ||
== Ongoing Topics == | == Ongoing Topics == | ||