Phonocardiogram (PCG) analysis is vital for cardiovascular disease diagnosis, yet the scarcity of labeled pathological data hinders the capability of AI systems. To bridge this, we introduce H-LDM, a Hierarchical Latent Diffusion Model for generating clinically accurate and controllable PCG signals from structured metadata. Our approach features: (1) a multi-scale VAE that learns a physiologically-disentangled latent space, separating rhythm, heart sounds, and murmurs; (2) a hierarchical text-to-biosignal pipeline that leverages rich clinical metadata for fine-grained control over 17 distinct conditions; and (3) an interpretable diffusion process guided by a novel Medical Attention module. Experiments on the PhysioNet CirCor dataset demonstrate state-of-the-art performance, achieving a Fréchet Audio Distance of 9.7, a 92% attribute disentanglement score, and 87.1% clinical validity confirmed by cardiologists. Augmenting diagnostic models with our synthetic data improves the accuracy of rare disease classification by 11.3%. H-LDM establishes a new direction for data augmentation in cardiac diagnostics, bridging data scarcity with interpretable clinical insights.

Current AI in high-stakes health monitoring like cardiology suffers from opacity and an inability to integrate established medical knowledge. We present Heuristica, a distributed neuro-symbolic system designed to address these limitations by providing transparent, interpretable reasoning for clinical review. Its core is a novel, sub- 100 ms bidirectional feedback loop between a Neural Perception Engine (NPE) and a Probabilistic Reasoning Graph (PRG), featuring a sub- 100 ms core reasoning cycle. This allows the symbolic PRG to not only interpret but actively guide the NPE's perception, fusing data-driven learning with medical logic in real-time. The system also features a privacy-preserving federated discovery protocol for aggregating relational insights across users without sharing raw data. Evaluations on cardiovascular benchmarks show significant accuracy gains ($+15.3 \%$F1-score), and promising clinical acceptance$(89.2 \%)$in a study with cardiologists, and can identify complex cardiovascular risk patterns validated on large-scale retrospective data. By creating a symbiotic loop between neural perception and symbolic reasoning, Heuristica provides a blueprint for the next generation of trustworthy and interpretable AI for ubiquitous health systems.

Federated Learning (FL) is vulnerable to model poisoning, where adversaries submit malicious updates to compromise the global model. A key challenge for defenses is the ambiguity between malicious updates and benign outliers from clients with non-identically distributed (Non-IID) data, forcing a trade-off between security and utility. We propose FedGuard, a novel defense framework that verifies the integrity of the local training process. FedGuard requires clients to submit a privacy-preserving proof of their training, encapsulated in a loss trajectory. A two-stage defense at the server filters updates with invalid dynamics and then uses a trajectory-aware anomaly detection algorithm to distinguish attacks from benign Non-IID outliers. Evaluations against backdoor attacks on MNIST and CIFAR-10 show FedGuard reduces the attack success rate to near-zero, significantly outperforming traditional defenses while maintaining high main task accuracy.

Serverless computing's performance is hampered by cold start latency, a barrier for latency-sensitive applications. Existing keep-alive strategies are costly, while platform optimizations are application-agnostic. We propose CoWarm, an application-aware framework leveraging function structural/temporal relationships via Chain Warming (call graphbased downstream warming) and Collaborative Warming (crossapplication co-occurrence mining). A cost-aware reinforcement learning (RL) policy balances latency and cost. Evaluations on a Kubernetes-based platform show Co-Warm reduces 99th percentile end-to-end latency by up to$\mathbf{4. 5 x}$, at$\mathbf{1 5 - 2 0 \%}$of naive keep-alive costs.