The $β$-VAE is a foundational framework for unsupervised disentanglement, using $β$ to regulate the trade-off between latent factorization and reconstruction fidelity. Empirically, however, disentanglement performance exhibits a pervasive non-monotonic trend: benchmarks such as MIG and SAP typically peak at intermediate $β$ and collapse as regularization increases. We demonstrate that this collapse is a fundamental information-theoretic failure, where strong Kullback-Leibler pressure promotes marginal independence at the expense of the latent channel's semantic informativeness. By formalizing this mechanism in a linear-Gaussian setting, we prove that for $β> 1$, stationarity-induced dynamics trigger a spectral contraction of the encoder gain, driving latent-factor mutual information to zero. To resolve this, we introduce the $λβ$-VAE, which decouples regularization pressure from informational collapse via an auxiliary $L_2$ reconstruction penalty $λ$. Extensive experiments on dSprites, Shapes3D, and MPI3D-real confirm that $λ> 0$ stabilizes disentanglement and restores latent informativeness over a significantly broader range of $β$, providing a principled theoretical justification for dual-parameter regularization in variational inference backbones.

Machine Unlearning aims to remove specific data from trained models, addressing growing privacy and ethical concerns. We provide a theoretical analysis of a simple and widely used method - gradient ascent - used to reverse the influence of a specific data point without retraining from scratch. Leveraging the implicit bias of gradient descent towards solutions that satisfy the Karush-Kuhn-Tucker (KKT) conditions of a margin maximization problem, we quantify the quality of the unlearned model by evaluating how well it satisfies these conditions w.r.t. the retained data. To formalize this idea, we propose a new success criterion, termed \textbf{$(ε, δ, τ)$-successful} unlearning, and show that, for both linear models and two-layer neural networks with high dimensional data, a properly scaled gradient-ascent step satisfies this criterion and yields a model that closely approximates the retrained solution on the retained data. We also show that gradient ascent performs successful unlearning while still preserving generalization in a synthetic Gaussian-mixture setting.

Deep learning has achieved tremendous success.

We revisit the replica method for analyzing inference and learning in parametric models, considering situations where the data-generating distribution is unknown or analytically intractable. Instead of assuming idealized distributions to carry out quenched averages analytically, we use a variational Gaussian approximation for the replicated system in grand canonical formalism in which the data average can be deferred and replaced by empirical averages, leading to stationarity conditions that adaptively determine the parameters of the trial Hamiltonian for each dataset. This approach clarifies how fluctuations affect information extraction and connects directly with the results of mathematical statistics or learning theory such as information criteria. As a concrete application, we analyze linear regression and derive learning curves. This includes cases with real-world datasets, where exact replica calculations are not feasible.