目次 at IgMin Research
We seek to promote cross-disciplinary innovation for quicker and more impactful outcomes.
Article Explorer
Malliavin Calculus as Stochastic Backpropagation for Gaussian Latent Models: A Variance-Optimal Hybrid Framework
Though Pancreatic Cancer may be Resistant to Immunotherapy from Immune Checkpoint Inhibitors, it may be Sensitive to Blocking the Production of Immunoprotective Proteins Unique to Pregnancy
Unmyelinated Aortic Baroreceptor Activity is Reduced in a Rabbit Model of Atherosclerosis
Research Article
Malliavin Calculus as Stochastic Backpropagation for Gaussian Latent Models: A Variance-Optimal Hybrid Framework
We establish a rigorous connection between pathwise (reparameterization) and score-function (Malliavin) gradient estimators by showing that both arise from the Malliavin integration-by-parts identity. Building on this equivalence, we introduce a unified and variance-aware hybrid estimator that adaptively combines pathwise and Malliavin gradients using their empirical covariance structure. The connection is established explicitly for Gaussian (and more generally exponential family) latent variable models, where integration-by-parts identities admit closed-form representations. The resulting formulation provides a principled understanding of stochastic backpropagation and achieves minimum variance in theory among all unbiased linear combinations, with closed-form finite-sample convergence bounds. We demonstrate 9% variance reduction on VAEs (CIFAR-10) and up to 35% on strongly-coupled synthetic problems. Exploratory policy gradient experiments reveal that non-stationary optimization landscapes present challenges for the hybrid approach, highlighting important directions for future work. Overall, this work positions Malliavin calculus as a conceptually unifying and practically interpretable framework for stochastic gradient estimation, clarifying when hybrid approaches provide tangible benefits and when they face inherent limitations.
