Search for: All records

The Probably Approximately Correct (PAC) Privacy framework [56] provides a powerful instance-based methodology to preserve privacy in complex data-driven systems. Existing PAC Privacy algorithms (we call them Auto-PAC) rely on a Gaussian mutual information upper bound. However, we show that the upper bound obtained by Auto-PAC is tight if and only if under the data distribution, the unperturbed output is Gaussian and the noise is independent Gaussian. We propose two approaches for addressing this issue. First, we introduce two tractable post-processing methods for Auto-PAC, based on Donsker–Varadhan representation and sliced Wasserstein distances. However, the result still leaves "wasted" privacy budget. To address this issue more fundamentally, we introduce Residual-PAC (R-PAC) Privacy, an f -divergence-based measure to quantify privacy that remains after adversarial inference. To implement R-PAC Privacy in practice, we propose a Stackelberg Residual-PAC (SR-PAC) automatic privatization algorithm, a game-theoretic framework that selects optimal noise distributions through convex bilevel optimization. Our approach achieves efficient privacy budget utilization for arbitrary data distributions and naturally composes when multiple mechanisms access the dataset. Our experiments demonstrate that SR-PAC obtains consistently a better privacy-utility tradeoff than both PAC and differential privacy baselines. 

We study the design of a privatization mechanism and privacy accounting in the Pufferfish Privacy (PP) family. Specifically, motivated by the curse of dimensionality and lack of practical composition tools for iterative learning in the recent Rényi Pufferfish Privacy (RPP) framework, we propose Sliced Rényi Pufferfish Privacy (SRPP). SRPP preserves PP/RPP semantics (customizable secrets with probability-aware secret–dataset relationships) while replacing high-dimensional Rényi divergence with projection-based quantification via two sliced measures, Average SRPP and Joint SRPP. We develop sliced Wasserstein mechanisms, yielding sound SRPP certificates and closed-form Gaussian noise calibration. For iterative learning systems, we introduce an SRPP-SGD scheme with gradient clipping and new accountants based on History-Uniform Caps (HUC) and a subsampling-aware variant (sa-HUC), enabling decompose-then-compose privatization and additive composition under a common slicing geometry. Experiments on static and iterative privatization show that the proposed framework exhibits favorable privacy–utility trade-offs, as well as practical scalability. 

Understanding how cells escape from embedded spheroids requires a mechanical framework that connects stress generation within cells, across cells, and between cells and the surrounding extracellular matrix (ECM). Here, we develop such a framework using a 3D vertex model of a spheroid coupled to a fibrous ECM network and derive the 3D Cauchy stress tensor for deformable polyhedral cells, enabling direct quantification of cell-level stresses in three dimensions beyond shape-based or bulk rheological measures. We characterize maximum shear stress distributions for solid-like and fluid-like spheroids. Solid-like spheroids exhibit broad stress distributions and spatial stress gradients, whereas fluid-like spheroids display lower stresses and minimal spatial patterning. We find little generic alignment between cell shape anisotropy and principal stress directions, demonstrating that morphology alone is not a reliable indicator of mechanical state. We further show that individual cells undergo strain stiffening upon elongation, producing nonlinear increases in maximum shear stress. This provides a mechanism by which boundary cells in otherwise low-stress, fluid-like spheroids can transiently generate forces sufficient to remodel the surrounding matrix. These results indicate that invasion cannot be understood solely through tissue-level unjamming transitions, but also involves cell-level stress amplification processes. To investigate how this strain-induced stress amplification couples to collective invasion modes, we introduce an extended 3D vertex model in which cells interact through explicit, tunable cell-cell adhesion springs. Within this minimal mechanical framework, single-cell breakout arises from the combination of strain stiffening and weakened cell-cell adhesion, whereas multi-cell streaming requires an additional ingredient: anisotropic adhesion strengthened along the axis of elongation in highly strained cells and weakened orthogonally. These findings identify distinct mechanical pathways linking cell strain, stress amplification, and adhesion organization to different modes of spheroid invasion. 

Abstract Intermediate filaments (IF) are diverse and cell-type specific. The IF protein vimentin, expressed in mesenchymal cells and different cancer cells, is functionally associated with cell migration through fibrous tissues. Vimentin increases cell elongation needed for migration, yet also acts as a cytoskeletal cage that hinders cells squeezing through small spaces. To determine how vimentin facilitates cell migration through the extracellular matrix (ECM) and around neighboring cells in tissues, we examine the collective invasion of cell spheroids embedded in collagen networks. Unlike single-cell migration in collagen networks, the collective invasion of cells through the collagen network is increased by vimentin for both mouse embryonic fibroblasts (MEF) and co-cultures of MEF with MDA-MB-231 breast cancer cells. Using multiple experimental systems, we show that vimentin increases spheroid contractility and that vimentin-mediated collective cell expansion depends on matrix metalloproteinases (MMP), which degrade collagen networks. In addition, through advanced imaging and a computational 3D cell vertex model, we find that vimentin enhances the elongation of cells in spheroids embedded in collagen, indicating increased spheroid fluidity and active collagen contraction. Altogether, these results reveal new insights on vimentin’s effects in enhancing collective cell migration in 3D matrix environments through collagen remodeling and tissue fluidity. 

Abstract Embedding a collective of tumor cells, i.e. a tumor spheroid, in a fibrous environment, such as a collagen network, provides an essentialin vitroplatform to investigate the biophysical mechanisms of tumor invasion. To predict new mechanisms, we develop a three-dimensional computational model of an embedded spheroid using a vertex model, with cells represented as deformable polyhedrons, mechanically coupled to a fiber network via active linker springs. As the linker springs actively contract, the fiber network remodels. As we tune the rheology of the spheroid and the fiber network stiffness, we find that both factors affect the remodeling of the fiber network with fluid-like spheroids densifying and radially realigning the fiber network more on average than solid-like spheroids but only for a range of intermediate fiber network stiffnesses. Our predictions are supported by experimental studies comparing non-tumorigenic MCF10A spheroids and malignant MDA-MB-231 spheroids embedded in collagen networks. The spheroid rheology-dependent effects are the result of cellular motility generating spheroid shape fluctuations. These shape fluctuations lead to emergent feedback between the spheroid and the fiber network to further remodel the fiber network. This emergent feedback occurs only at intermediate fiber network stiffness since at low fiber network stiffness, the mechanical response of the coupled system is dominated by the spheroid and for high fiber network stiffness, the mechanical response is dominated by the fiber network. We are therefore able to quantify the regime of optimal spheroid-fiber network mechanical reciprocity. Our results uncover intricate morphological-mechanical interplay between an embedded spheroid and its surrounding fiber network with both spheroid contractile strengthandspheroid shape fluctuations playing important roles in the pre-invasion stages of tumor invasion. 

Our new multiscale model proposes a critical nuclear strain that triggers ZEB2 activation, delayed in human-derived organoids. 

Increasing studies have shown bugs in multi-language software as a critical loophole in modern software quality assurance, especially those induced by language interactions (i.e., multilingual bugs). Yet existing tool support for bug detection/localization remains largely limited to single-language software, despite the long-standing prevalence of multi-language systems in various real-world software domains. Extant static/dynamic analysis and deep learning (DL) based approaches all face major challenges in addressing multilingual bugs. In this paper, we present xLoc, a DL-based technique/tool for detecting and localizing multilingual bugs. Motivated by results of our bug-characteristics study on top locations of multilingual bugs, xLoc first learns the general knowledge relevant to differentiating various multilingual control-flow structures. This is achieved by pre-training a Transformer model with customized position encoding against novel objectives. Then, xLoc learns task-specific knowledge for the task of multilingual bug detection/localization, through another new position encoding scheme (based on cross-language API vicinity) that allows for the model to attend particularly to control-flow constructs that bear most multilingual bugs during fine-tuning. We have implemented xLoc for Python-C software and curated a dataset of 3,770 buggy and 15,884 non-buggy Python-C samples, which enabled our extensive evaluation of xLoc against two state-of-the-art baselines: fine-tuned CodeT5 and zero-shot ChatGPT. Our results show that xLoc achieved 94.98% F1 and 87.24%@Top-1 accuracy, which are significantly (up to 162.88% and 511.75%) higher than the baselines. Ablation studies further confirmed significant contributions of each of the novel design elements in xLoc. With respective bug-location characteristics and labeled bug datasets for fine-tuning, our design may be applied to other language combinations beyond Python-C. 

Multi-axial real-time hybrid simulation (maRTHS) uses multiple hydraulic actuators to apply loads and deform experimental substructures, enacting bothtranslationalandrotationalmotion. This allows for an increased level of realism in seismic testing. However, this also demands the implementation of multiple-input, multiple-output control strategies with complex nonlinear behaviors. To realize true real-time hybrid simulation at the necessary sub-millisecond timescales, computational platforms will need to support these complexities at scale, while still providing deadline assurance. This paper presents initial work towards supporting (and is influenced by the need for) envisioned larger-scale future experiments based on the current maRTHS benchmark: it discusses aspects of hardware, operating system kernels, runtime middleware, and scheduling theory that may be leveraged or developed to meet those goals. This work aims to create new concurrency platforms capable of managing task scheduling and adaptive event handling for computationally intensive numerical simulation and control models like those for the maRTHS benchmark problem. These should support real-time behavior at millisecond timescales, even for large complex structures with thousands of degrees of freedom. Temporal guarantees should be maintained across behavioral and computational mode changes, e.g., linear to nonlinear control. Pursuant to this goal, preliminary scalability analysis is conducted towards designing future maRTHS experiments. The results demonstrate that the increased capabilities of modern hardware architectures are able to handle larger finite element models compared to prior work, while imposing the same latency constraints. However, the results also illustrate a subtle challenge: with larger numbers of CPU cores, thread coordination incurs more overhead. These results provide insight into the computational requirements to support envisioned future experiments that will take the maRTHS benchmark problem to nine stories and beyond in scale. In particular, this paper (1) re-evaluates scalability of prior work on current platform hardware, and (2) assesses the resource demands of a basic smaller scale model from which to gauge the projected scalability of the new maRTHS benchmark as ever larger and more complex models are integrated within it.