Code and experiments

Personalised federated graph learning with topology-weighted aggregation, a proximal term and degree-calibrated differential privacy, benchmarked on seven graph datasets with 10 and 20 clients.

How the language of the calibration data shapes 4-bit GPTQ and AWQ quantization of multilingual LLMs, with a Turkish focus and a tokenizer-density analysis.

A PyTorch Geometric framework for graph, node and link tasks with Graph Trojaning Attack backdoors on graph classification, plus spectral, topological and feature analyses of poisoned graphs.

Corruption and adversarial robustness of transfer-learned and distilled CNNs on CIFAR-10/-C: AugMix with JSD, PGD, Grad-CAM, and how a teacher's adversarial examples transfer to its students.

End-to-end learned channel codes for the AWGN channel with noiseless feedback: a transformer transmitter and receiver trained through the channel, with ablations over rounds, noise and capacity.

Product recommendation from Amazon reviews via instruction-tuned aspect-based sentiment analysis, LDA and BERTopic topics, and hierarchical meta-aspect clustering.

A plug-and-play library of evasion attacks, backdoor attacks and defenses, and Byzantine-robust federated learning on MNIST, with one runnable script per topic.

Style-conditioned Turkish LLMs: retrieval over a columnist's articles, LoRA fine-tuning, and an evaluation of style imitation against faithfulness to the source text.

A 22-run ablation of depth, width, activation, dropout, BatchNorm and L1/L2 regularisation for an MLP on MNIST, orchestrated with SLURM.