View source for AI-Driven Semantic Encoding for Efficient Communication

{{StudentProjectTemplate
|Summary=To develop and evaluate an AI-based semantic encoding model capable of transforming raw data into compact, structured representations. 
|References=1) Pezone, Francesco. "Semantic communication based on generative AI: a new approach to image compression and edge optimization." arXiv preprint arXiv:2502.01675 (2025).

2) Qiao, Li, Mahdi Boloursaz Mashhadi, Zhen Gao, Chuan Heng Foh, Pei Xiao, and Mehdi Bennis. "Latency-aware generative semantic communications with pre-trained diffusion models." IEEE wireless communications letters (2024).

3) Islam, Azharul, and KyungHi Chang. "Navigating the future of wireless networks: A multidimensional survey on semantic communications." ICT Express 10, no. 4 (2024): 747-773.

4) Li, Nan, Alexandros Iosifidis, and Qi Zhang. "Dynamic semantic compression for cnn inference in multi-access edge computing: A graph reinforcement learning-based autoencoder." IEEE Transactions on Wireless Communications (2024).
|Supervisor=EDISON PIGNATON DE FREITAS
|Level=Master
|Status=Open
}}
Contextualization: 
Semantic Communication Networks represent a fundamental shift in how we think about and design communication systems. Unlike traditional communication, which focuses on the accurate and reliable transmission of bits, SCNs prioritize the meaning or intent behind the data.

One of the key element in SCN is the semantic encoder. Instead of compressing data using a general-purpose algorithm (like JPEG for images), a sophisticated AI model analyzes the raw data and extracts only the most important semantic features. The encoder compresses this semantic information into a low-dimensional vector that captures the properties of the original data.

Goals:
To develop and evaluate an AI-based semantic encoding model capable of transforming raw data into compact, structured representations. The objective is to reduce transmission payload while preserving the essential meaning of the information. The proposed encoder will focus on text sentences and/or images as primary data modalities.

Testing; 
Datasets recommendation: European Parliament for text and MNIST for images.

Evaluation Criteria:
Compare the proposed solution with traditional communication (the baseline) in terms of compression ratio; semantic similarity; CPU/GPU consumption and encoding/decoding time.

Extra (if you have time): behavior/evaluation when a normal noise is inserted in both baseline and proposed solution.

Main Tasks: 
Task 1: Study semantic encoders concepts, especially NLPs, LLMs, ANNs Knowledge Graphs and Topos. 
Task 2: Study the addressed dataset. 
Task 3: Develop, train and improve the model. 
Task 4: Test the model, capturing the evaluation criteria.  
Task 5: Analisys the results and compare with baseline. 
Task 6: Write thesis and prepare presentation. 

Deliverables (Besides the final thesis document): 
- AI model specification to encode and decode text/image. 
- Tests description with visual result representation (graphs).
- Performance analysis report. 
- Thesis documentation.