Chalmers Open Digital Repository

Välkommen till Chalmers öppna digitala arkiv!

Här hittar du:

  • Studentarbeten utgivna på lärosätet, såväl kandidatarbeten som examensarbeten på grund- och masternivå
  • Digitala specialsamlingar, som t ex Chalmers modellkammare
  • Utvalda projektrapporter

Forskningspublikationer, rapporter och avhandlingar hittar du i research.chalmers.se

 

Enheter i Chalmers ODR

Välj en enhet för att se alla samlingar.

Senast inlagda

Bild (thumbnail)
Bild (thumbnail)
Bild (thumbnail)
PRED-RAG: a Predictive Radial Grid for Automotive Radar Multipath - Identification Identification of objects created by the radar multipath phenomenon, with focus on low computational complexity.
(2025) Kindlund, Erik; Karlsson, Andreas
Automotive radar sensors are crucial for advanced driver assistance systems but are susceptible to the multipath phenomenon, where radio waves reflect multiple times between surfaces, creating false "ghost" objects that can trigger unnecessary safety interventions. Previous work relies on restrictive assumptions about reflection surfaces and environmental conditions, yielding solutions that perform well in specific scenarios but demonstrate limited generalization capabilities in the complex, diverse situations encountered during real-world driving. This thesis addresses the challenge of identifying radar multipath objects in real-time environments, focusing on developing an algorithm that maintains low computational complexity while achieving high accuracy. We established a development and evaluation pipeline using synthetic data together with a simulation framework, enabling data driven development of our algorithm. We propose the PRED-RAG algorithm, a novel approach that utilizes a radial grid structure combined with host motion prediction of static detections for enhanced high-level environment mapping. The algorithm identifies triplets consisting of a ghost object, reflection point and true object, then evaluates them using velocity-based criteria. When compared to a state-of-the-art algorithm, our approach demonstrates superior performance in both accuracy and computational efficiency across various driving scenarios. The PRED-RAG algorithm achieves 94.43% accuracy for high-priority objects compared to 39.26% for the baseline, with significantly better generalization capabilities, particularly in complex environments. The geometric properties employed in the grid-based approach effectively separate ghost objects from true objects while maintaining runtime performance suitable for real-time automotive applications. This work contributes to safer autonomous driving systems by reducing false objects that could lead to unnecessary emergency interventions.
Bild (thumbnail)
Methods for Optimizing BERT Model on Edge Devices - Accelerating Biomedical NLP with Pruned and Quantized BERT Models
(2025) Barani, Amir Ali; Mirzabeigi, Atefeh
Named-entity recognition (NER) of clinical efficacy endpoints in oncology abstracts supports downstream discovery pipelines at AstraZeneca. Yet, the fine-tuned transformer models currently used are too slow and over parameterized for large-scale CPU deployment. This thesis evaluated whether post-training model compression techniques can accelerate inference without retraining or harming extraction quality. In the first stage of this project, standard BERT and BioBERT were individually pruned with a three-stage, Fisher-guided structured pruning workflow at three levels of sparsity. Subsequently, in the second stage, dynamic 8-bit integers quantization using ONNX Runtime was applied to standard BERT, BioBERT, and DistilBERT. The third stage involved combining both pruning and quantization, further optimizing the pre-trained standard BERT and BioBERT transformers. Experiments were run on annotated MEDLINE sentences covering 25 efficacy labels, with F1 score and inference latency per sample serving as primary metrics. A 25% structured-sparsity level yielded no measurable drop in F1 score, and the additional 8-bit integers dynamic step cut latency further. The best configuration, 25%- pruned+8-bit integers BioBERT, reduced mean CPU inference time from 32.52 ms to 12.02 ms (2.6-fold speed-up) while accuracy fell only from 0.982 to 0.980 and F1 score from 0.954 to 0.948. The Post-training structured pruning combined with 8-bit integers dynamic quantization makes the oncology-NER pipeline about three times faster in inference time on standard CPUs without compromising the extraction quality or needing special hardware or libraries.
Bild (thumbnail)
Mobile Traffic Classification Over VPN - Evaluating Encrypted Traffic Classification Techniques on VPN Traffic: A Comparative Study
Ghalayini, Hassan; Yifter, Nahusenay
In recent years, mobile network traffic classification has gained significant attention from network operators to better understand customer needs and allocate bandwidth based on application requirements. Research on machine learning and deep learning models has increased in popularity, as these methods enable more accurate classification while leveraging different aspects of network packets beyond just the payload. The primary goal of this thesis is to compare the performance of different state-of-theart deep learning models namely, Convolutional Neural Networks (CNNs), Recurrent Neural Networks(RNNs), and Autoencoders(AEs) through a series of experiments and to evaluate the feasibility of deploying these models for network classification for mobile traffic in VPN environments. The study focuses on network packets that are both encrypted and tunneled over Virtual Private Networks (VPN). A dataset of 50GB of VPN data is used to train, assess, and enhance analysis and training of the models. Our results indicate that CNNs effectively extract features but struggle with capturing sequential dependencies. By comparison, RNNs demonstrate greater efficiency in recognising temporal patterns and achieve higher recall rates. Autoencoders perform well for specific application classes but exhibit lower precision and recall overall. This thesis suggests further investigation into a combined approach between convolutional neural networks and recurrent neural networks to be used for traffic classification over VPNs.