Hansung Choi

Hansung Choi, Daewon Seo

Recent advances in deep learning-based joint source-channel coding (deepJSCC) have substantially improved communication performance, but their high computational cost hinders practical deployment. Moreover, certain applications require the ability to dynamically adapt computational complexity. To address these issues, we propose a Feature Importance-Aware deepJSCC (FAJSCC) model for image transmission that is both computationally efficient and adjustable. FAJSCC employs axis-dimension specialized computation, which performs efficient operations individually for each spatial and channel axis, significantly reducing computational cost while representing features effectively. It further incorporates selective deformable self-attention, which applies self-attention only to selected and adaptively adjusted features, leveraging the importance and relations of input features to efficiently capture complex feature correlations. Another key feature of FAJSCC is that the number of selected important areas can be controlled separately by the encoder and the decoder, depending on the available computational budget. It makes FAJSCC the first deepJSCC architecture to allow independent adjustment of encoder and decoder complexity within a single trained model. Experimental results show that FAJSCC achieves superior image transmission performance under various channel conditions while requiring less computational complexity than recent state-of-the-art models. Furthermore, experiments independently varying the encoder and decoder's computational resources reveal, for the first time in the deepJSCC literature, that understanding the meaning of noisy features in the decoder demands the greatest computational cost. The code is publicly available at github.com/hansung-choi/FAJSCCv2.

Hansung Choi, Daewon Seo

The concept of a minimax classifier is well-established in statistical decision theory, but its implementation via neural networks remains challenging, particularly in scenarios with imbalanced training data having a limited number of samples for minority classes. To address this issue, we propose a novel minimax learning algorithm designed to minimize the risk of worst-performing classes. Our algorithm iterates through two steps: a minimization step that trains the model based on a selected target prior, and a maximization step that updates the target prior towards the adversarial prior for the trained model. In the minimization, we introduce a targeted logit-adjustment loss function that efficiently identifies optimal decision boundaries under the target prior. Moreover, based on a new prior-dependent generalization bound that we obtained, we theoretically prove that our loss function has a better generalization capability than existing loss functions. During the maximization, we refine the target prior by shifting it towards the adversarial prior, depending on the worst-performing classes rather than on per-class risk estimates. Our maximization method is particularly robust in the regime of a small number of samples. Additionally, to adapt to overparameterized neural networks, we partition the entire training dataset into two subsets: one for model training during the minimization step and the other for updating the target prior during the maximization step. Our proposed algorithm has a provable convergence property, and empirical results indicate that our algorithm performs better than or is comparable to existing methods. All codes are publicly available at https://github.com/hansung-choi/TLA-linear-ascent.