arXiv subjects

Zonghao Liu

Publications and source records attributed to Zonghao Liu.

8 recordsLinked to original sources

arXiv2026

FTU-Seek: Foundation Model-Guided Hard-Negative Learning for Sparse Functional Tissue Unit Segmentation

Functional tissue units (FTUs), including tertiary lymphoid structures (TLSs), blood vessels, and glands, encode localized immune, vascular, and epithelial organization in histopathology. Accurate quantification of these structures is important for studying tissue architecture and disease-associated tissue organization. However, FTUs are frequently sparse, heterogeneous, and surrounded by large amounts of morphologically similar background tissue, making automated segmentation in whole-slide images (WSIs) challenging. We therefore developed FTU-Seek, a pathology foundation model-guided framework that treats morphology-aware negative-patch selection as a key component of sparse FTU segmentation. FTU-Seek uses frozen multi-depth features from the UNI pathology foundation model to train a patch-level classifier that distinguishes FTU-containing from FTU-absent tissue. Target-absent patches are subsequently ranked according to their predicted target-containing probabilities, and the highest-scoring hard negatives are selected through a static Top$K$ strategy to construct compact segmentation training sets. The framework was evaluated using five-fold cross-validation and internal test cohorts across TLS, blood-vessel, and gland segmentation tasks, with an additional independent 30-WSI held-out cohort for TLS. Positive-only, all-tissue, random-negative, and matched random Top$K$ sampling strategies served as comparators. Segmentation-derived phenotypes were further explored in external TCGA cohorts.

cs.CV↗

arXiv2026

Rethinking AI Cloud Infrastructure for Agentic Serving Systems with the Aries Experimentation Framework

Autonomous agents challenge conventional LLM serving by coupling repeated inference with persistent context and sandboxed tool execution. We present Aries, a full-stack experimentation framework that separates task semantics from execution configurations, reconstructs cross-component agent trajectories with correlated system telemetry, and exposes stateful tool execution through a consistent interface across heterogeneous sandbox substrates. We use Aries to conduct reproducible experiments on open agent harnesses and benchmarks. We complement these experiments with production traces from a commercial platform, grounding low-level systems research in observed production behavior. Our results show that (1) token-centric metrics miss non-inference bottlenecks, (2) retaining additional context yields diminishing accuracy benefits while reducing serving capacity, and (3) tool sandboxes alternate between long idle periods and short resource bursts, while current snapshot-based state management makes aggressive suspension costly. A complementary security analysis further highlights the need to reduce the sandbox attack surface. We then discuss the vision for agent-native serving systems designed around trajectory-level metrics, adaptive context management, elastic sandbox resource management, and sandboxes with minimized attack surface.

cs.DC↗

arXiv2026

Computational Oncology of Chemotaxis-Driven Tumour--Immune Spatial Patterning and Stability

We develop a reaction--diffusion--chemotaxis model for spatial tumour--immune--chemokine dynamics that couples logistic tumour growth, immune-mediated killing, chemokine-dependent immune recruitment, chemotactic migration, and signal production. For the nondimensional system, we establish local classical solvability, nonnegativity, a uniform tumour-density bound, and global mass estimates for the immune and chemokine components. The tumour-free equilibrium is stable precisely when the baseline immune-control index satisfies $\sigma_0/\delta>1$, whereas positive homogeneous coexistence is characterized by a scalar nonlinear equation. Linearization in the Neumann Laplacian eigenbasis yields a mode-dependent cubic dispersion relation, showing that chemotaxis does not alter the tumour-invasion threshold but can destabilize homogeneous coexistence through a finite-wavelength oscillatory instability above a critical sensitivity $\xi_c$. A conservative finite-volume discretization with upwind chemotactic fluxes and implicit backward differentiation formula time integration is used to test these predictions. Numerical experiments recover the analytical equilibria and growth rates, identify the dominant unstable mode, reproduce the transition to spatial heterogeneity, and quantify the effects of immune recruitment, decay, and diffusion on the stability boundary. Grid-refinement, mass-balance, residual, and nonnegativity diagnostics support the computational reliability of the results.

math.AP↗

arXiv2026

Mode-Wise Spectral Criteria for Coupled Mass Transport in Hybrid PDE--ODE Tumor Microenvironments

We study coupled mass transport in a tumor--microenvironment setting with two motile densities $(S,R)$ and non-motile state switching $(P,A)$. The populations diffuse and undergo chemotactic drift; $(P,A)$ follow pointwise ODE switching. A decoupled inhibitory field $D$ satisfies a damped Neumann heat equation, giving maximum-principle bounds and exponential decay. Together with the pointwise invariant $P+A$, these identities yield global existence, positivity, and long-time reduction to limiting $(S,R)$ kinetics with a unique globally attracting coexistence state. Neumann eigenmode reduction gives closed dispersion relations. The base $(S,R)$ reaction--diffusion block remains stable for all nonconstant modes for any $d_S,d_R>0$, excluding classical Turing destabilization. Chemotaxis is posed via a diffusive cue $c$, since $\nabla A$ is undefined for non-diffusive $A$. In one-way damped coupling, the linearized mode matrix is block triangular and leaves the $(S,R)$ spectrum unchanged. Two-way coupling adds a feedback rank-one mobility correction, induces effective cross-diffusion, and admits mode growth. We give explicit trace/determinant criteria for unstable Laplacian modes and the resulting instability thresholds.

math.AP↗

arXiv2026

Chemotactic Feedback Controls Patterning in Hybrid Tumor--Stroma Model

Motivated by an ongoing collaboration with clinical oncologists and pathologists, we develop a hybrid partial differential equation--ordinary differential equation (PDE--ODE) framework that captures (i) competition between susceptible and resistant phenotypes, (ii) stromal state switching, and (iii) a clinically realistic open-loop, single-dose therapeutic agent $I$ with diffusion and clearance. Clinical management of solid tumors is increasingly limited by spatial heterogeneity and therapy-induced resistance niches that are difficult to predict from well-mixed models. We establish a rigorous mathematical backbone with forward invariance of the nonnegative cone and global-in-time well-posedness. Exploiting the decoupled drug equation $\partial_t I=d_I\Delta I-\gamma_I I$, we prove a long-time reduction during washout and show that the damped base dynamics admit no diffusion-driven (Turing-type) instability. We then formulate a directionality--damping principle: unidirectional (open-loop) sensing yields at most transient focusing, whereas bidirectional (closed-loop) feedback reshapes the effective mobility and produces explicit thresholds separating stable homogeneity, finite-band patterning (resistance niche formation), and aggregation when strong parabolicity is violated. Reproducible simulations corroborate this classification and highlight when flux regularization is required for physical realism.

math.AP↗

arXiv2025

TokenScale: Timely and Accurate Autoscaling for Disaggregated LLM Serving with Token Velocity

The architectural shift to prefill/decode (PD) disaggregation in LLM serving improves resource utilization but struggles with the bursty nature of modern workloads. Existing autoscaling policies, often retrofitted from monolithic systems like those in AIBrix and DistServe, rely on lagging indicators such as GPU utilization or coarse-grained request counts. This results in slow reactions to load spikes, leading to significant Time-to First-Token (TTFT) and Time-Per-Output-Token (TPOT) SLO violations and costly over-provisioning. We introduce TokenScale, an autoscaling framework that resolves this performance mismatch through two innovations. First, we propose Token Velocity, a novel metric that unifies the prefill, network, and decode stages by quantifying their rate of work. As a leading indicator of system backpressure, it enables proactive scaling. Second, Convertible Decoders allow decoder GPUs to dynamically execute prefill tasks during traffic spikes, creating a rapid-response buffer that absorbs bursts and eliminates the initialization latency of new prefillers. Our evaluation on a GPU cluster with production traces shows TokenScale improves SLO attainment from 50-88% to 80-96% and reduces costs by 4-14% over state-of-the-art systems, including DistServe, BlitzScale, and AIBrix. By uniting a predictive metric with a flexible system design, TokenScale significantly boosts the performance and efficiency of disaggregated LLM serving infrastructure.

cs.DC↗

arXiv2025

A hybrid PDE-ABM model for angiogenesis and tumour microenvironment with application to resistance in cancer treatment

The main obstacle to effective cancer treatment is the development of drug resistance, which can be divided into two categories: spontaneous and acquired drug resistance. Non-small cell lung cancer (NSCLC) is the main cause of cancer-related deaths worldwide. A subset of lung cancer, adenocarcinomas, is characterised by mutations in the epidermal growth factor receptor (EGFR) gene. Treatment of EGFR-mutated lung adenocarcinomas has become less effective over time due to drug resistance development, which is associated with a second mutation in the EGFR gene. An important factor in the development of cancer is angiogenesis, which is the formation of blood vessels from the existing vasculature. These newly formed blood vessels provide oxygen and nutrients to tumour cells to maintain tumour growth and proliferation. We applied a hybrid discrete-continuous (HDC) model to capture the dynamic vasculature in the tumour microenvironment (TME). In the case of pre-existing resistance, the formation of angiogenic networks creates a microenvironment that supports tumour survival and enhances drug resistance. In the case of spontaneous mutation-induced resistance, earlier and more frequent mutations confer a greater survival advantage to the tumour population. There is also a mutually reinforcing relationship between a high proliferation rate and high resistance characteristics. These findings explain two conflicting experimental results about the second mutation in NSCLC.

math.NA↗

arXiv2024

Time-resolved optical assessment of exciton formation in mixed two-dimensional perovskite films

We report the observation of exciton formation from the cooled band-edge carriers in mixed two-dimensional hybrid organic-inorganic perovskites using femtosecond transient absorption spectroscopy. By monitoring the changes of bleach signal upon excitations with various photon energy, we are able to extract the values of exciton binding energy and the occupancies of carriers of free and bound states for each two-dimensional phase. We also confirm the existence of Mahan exciton when injected carrier density is above the Mott criterion.

physics.optics↗