CRoP: Context-wise Robust Static Human-Sensing Personalization

TL;DR

CRoP tackles intra-user generalization under distribution shifts in static personalization for human sensing by combining adaptive pruning with a mixing step that integrates generic knowledge from off-the-shelf pretrained models. It learns a user-specific subnetwork from limited available-context data and restores generic signals through a mixing operation, enabling robust performance on unseen contexts without accessing them during training. Across four real-world health-sensing datasets, CRoP yields substantial personalization gains and maintains positive generalization relative to baselines, supported by gradient inner-product analyses and ablations. The method is architecture-agnostic and on-device friendly, offering privacy-preserving, context-aware personalization for clinical and health-monitoring applications.

Abstract

The advancement in deep learning and internet-of-things have led to diverse human sensing applications. However, distinct patterns in human sensing, influenced by various factors or contexts, challenge the generic neural network model's performance due to natural distribution shifts. To address this, personalization tailors models to individual users. Yet most personalization studies overlook intra-user heterogeneity across contexts in sensory data, limiting intra-user generalizability. This limitation is especially critical in clinical applications, where limited data availability hampers both generalizability and personalization. Notably, intra-user sensing attributes are expected to change due to external factors such as treatment progression, further complicating the challenges. To address the intra-user generalization challenge, this work introduces CRoP, a novel static personalization approach. CRoP leverages off-the-shelf pre-trained models as generic starting points and captures user-specific traits through adaptive pruning on a minimal sub-network while allowing generic knowledge to be incorporated in remaining parameters. CRoP demonstrates superior personalization effectiveness and intra-user robustness across four human-sensing datasets, including two from real-world health domains, underscoring its practical and social impact. Additionally, to support CRoP's generalization ability and design choices, we provide empirical justification through gradient inner product analysis, ablation studies, and comparisons against state-of-the-art baselines.

Figures (8)

• Figure 1: Problem Setting: Study Objective
• Figure 2: Model Structure for the Lenet Model used for Widar Dataset. The parameters of the highlighted layer are used for preliminary analysis
• Figure 3: Heat map for the absolute magnitude of parameters belonging to penultimate layer for LeNet models finetuned using data from context (a) C1 and (b) C2. The first linear layer after the convolution layer has been used for this analysis. The size of the original weight matrix of the layer is 1536 X 128. However, we reshape heat map to 384 X 512 for better observability.
• Figure 4: Detailed results for Stress sensing dataset User 3 (Scenario 1) through F1 score
• Figure 5: t-SNE plots for user '7c' in ExtraSensory Dataset for Scenario 2 where avalable context $\mathcal{C}_a$ ('phone in pocket' and 'phone in bag') and $\mathcal{C}_u$ ('phone in hand')context. The colors of the samples indicate their class membership.