COMP6258: Differentiable Programming and Deep Learning

2025-26


Maintained by Professor Jonathon Hare and Dr Antonia Marcu.

Deep Learning Coursework: The COMP6258 Reproducibility Challenge - Paper Ideas

Title URL
Activation by Interval-wise Dropout A Simple Way to Prevent Neural Networks from Plasticity Loss https://arxiv.org/pdf/2502.01342
DESIGNING CONCISE CONVNETS WITH COLUMNAR STAGES https://proceedings.iclr.cc/paper_files/paper/2025/file/a12e362d89d4e0b40760f839f91550ee-Paper-Conference.pdf
Improving Out-of-Distribution Detection via Dynamic Covariance Calibration https://openreview.net/pdf?id=UjLxG9k4B6
Collapse-Proof Non-Contrastive Self-Supervised Learning https://openreview.net/pdf?id=wIfl8PK6Op
ABKD: Pursuing a Proper Allocation of the Probability Mass in Knowledge Distillation via α-β-Divergence https://openreview.net/pdf?id=vt65VjJakt
Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations https://openreview.net/pdf?id=TPuFRuNano
Are Sparse Autoencoders Useful? A Case Study in Sparse Probing https://openreview.net/pdf?id=rNfzT8YkgO
AuroRA: Breaking Low-Rank Bottleneck of LoRA with Nonlinear Mapping https://openreview.net/pdf?id=2hgHyoyVWj
Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning https://arxiv.org/pdf/2502.07560
NegMerge: Sign-Consensual Weight Merging for Machine Unlearning https://arxiv.org/pdf/2410.05583
Predicting the Susceptibility of Examples to Catastrophic Forgetting https://arxiv.org/pdf/2406.09935
BOOD: Boundary-based Out-Of-Distribution Data Generation https://arxiv.org/pdf/2508.00350
Weakly-Supervised Contrastive Learning for Imprecise Class Labels https://arxiv.org/pdf/2505.22028
Sum-of-Parts: Self-Attributing Neural Networks with End-to-End Learning of Feature Groups https://arxiv.org/pdf/2310.16316
Quamba2: A Robust and Scalable Post-training Quantization Framework for Selective State Space Models https://openreview.net/pdf?id=Zm0Kper4yx
Synergy Between the Strong and the Weak: Spiking Neural Networks are Inherently Self-Distillers https://openreview.net/pdf?id=BrmR69AhUg
Spiking Neural Networks Need High-Frequency Information https://openreview.net/pdf?id=owNPAl7LNK
Pixel2Feature Attack (P2FA): Rethinking the Perturbed Space to Enhance Adversarial Transferability https://openreview.net/pdf?id=bPJo5uSkOJ
NeuronTune: Towards Self-Guided Spurious Bias Mitigation https://openreview.net/pdf?id=qC5FZs34Xr
Copyright ©2025 The University of Southampton. All rights reserved.