Domain Generalization (DG) requires models to generalize across unseen data distributions. Kernel-based theory reveals a No-Free-Lunch problem: any model with a fixed representation is fundamentally sub-optimal for all possible shifts. While large ensembles mitigate this, they are computationally expensive and remain static once trained, inheriting the same theoretical limitation. We introduce MoE² (Mixture-of-Mixtures of Experts), a framework that uses a single frozen backbone to dynamically synthesize a bespoke adapter for each input, allowing it to continuously adapt its effective kernel. We provide a theoretical grounding for this process, proving our routing mechanism is a principled non-parametric estimator for the optimal Bayes mixture of experts. We derive a generalization bound that cleanly separates the router's estimation error from the reduction in a kernel-mismatch penalty achieved via synthesis. MoE² matches or exceeds state-of-the-art ensemble baselines on major DG benchmarks while using only a single, compact model. MoE² thus provides a theoretically-grounded and lightweight alternative to large-scale ensembles for robust domain generalization.

In this paper, we design a deep learning-based convolutional autoencoder for channel coding and modulation. The objective is to develop an adaptive scheme capable of operating at various signal-to-noise ratios (SNR)s without the need for re-training. Additionally, the proposed framework allows validation by testing all possible codes in the codebook, as opposed to previous AI-based encoder/decoder frameworks which relied on testing only a small subset of the available codes. This limitation in earlier methods often led to unreliable conclusions when generalized to larger codebooks. In contrast to previous methods, our multi-level encoding and decoding approach splits the message into blocks, where each encoder block processes a distinct group of B bits. By doing so, the proposed scheme can exhaustively test 2Bpossible codewords for each encoder/decoder level, constituting a layer of the overall scheme. The proposed model was compared to classical polar codes and TurboAE-MOD schemes, showing improved reliability with achieving comparable, or even superior results in some settings. Notably, the architecture can adapt to different SNRs by selectively removing one of the encoder/decoder layers without re-training, thus demonstrating flexibility and efficiency in practical wireless communication scenarios.

The paper provides a new perspective on peak- and average-constrained Gaussian channels. Such channels model optical wireless communication (OWC) systems which employ intensity-modulation with direct detection (IM/DD). First, the paper proposes a new, capacity-preserving vector binary channel (VBC) model, consisting of dependent binary noisy bit-pipes. Then, to simplify coding over this VBC, the paper proposes coding schemes with varying levels of complexity, building on the capacity of binary-symmetric channels (BSC) and channels with state. The achievable rates are compared to capacity and capacity bounds, showing that coding for the BSC with state over the VBC achieves rates close to capacity at moderate to high signal-to-noise ratio (SNR), whereas simpler schemes achieve lower rates at lower complexity. The presented coding schemes are realizable using capacity-achieving codes for binary-input channels, such as polar codes. Numerical results are provided to validate the theoretical results and demonstrate the applicability of the proposed schemes.