arXiv subjects

Zhizhuo Zhang

Publications and source records attributed to Zhizhuo Zhang.

15 recordsLinked to original sources

arXiv2026

Efficient Bezier Velocity Optimization for Free-Floating Space Manipulators

We present FAVOR (Free-floating Arm Velocity Optimization with Recursive Sensitivities), a planner for collision-free reaching, tracking, and prescribed-time pre-grasp interception on an unactuated spacecraft. It optimizes Bezier joint-velocity curves with linear velocity, acceleration, and continuity constraints. Decision dimension is independent of rollout resolution. Analytical recursive sensitivities provide task and clearance gradients through the coupled base-arm motion. Parallel evaluation, caching, and incremental collision discovery reduce computation. With a seven-DoF arm and five simulated spacecraft models, FAVOR achieves 99.8% point-to-point success with 1.779 s mean computation, versus 62.9% and 50.270 s for an IK-initialized position-spline baseline. Means include failures and timeouts. FAVOR completes 30 of 36 tracking cases, versus 15 for single-step QP, and all 36 interception instances, versus 22 for the spline baseline. A controlled ablation shows that finite differences increase mean planning time 4.7-fold.

cs.RO↗

arXiv2026

Improving Molecular-Morphology Contrastive Pretraining using Deep-Learning-based Morphology Profiles

Recent advancements in image-based profiling techniques have enabled the collection of high-volume cell morphology data, allowing new molecular embedding models to learn from the experimental phenotypic perturbations of a molecule in a cell. Previously, we developed Molecule-Morphology Contrastive Pretraining (MoCoP), a strategy for aligning small molecule embeddings to morphology fingerprints extracted through CellProfiler. The resulting molecular representation showed transferable performance for quantitative structure--activity relationship (QSAR) prediction tasks. Here, we extend the method by using a deep-learning-based cell image encoding pipeline to extract more feature-rich morphology profiles and align them to the molecular embeddings through contrastive learning. The new embeddings encode more accurate information on how molecules perturb cell morphology and enable improvements for QSAR predictions through either fixed-embedding linear probes or fully flexible fine-tuning. Morphology retrieval performance scales log-linearly with training data size, suggesting continued improvements as larger datasets become available. The improved MoCoP v2 also achieves superior performance on toxicity prediction and competitive results on ADME and activity benchmarks, when compared with existing molecular embedding models that use both cell morphology and transcriptomic data during training.

cs.LG↗

arXiv2026

Navigating heterogeneous protein landscapes through geometry-aware smoothing

The evolutionary fitness landscape of biological molecules is extremely sparse and heterogeneous, with functional sequences forming isolated dense ``islands'' within a vast combinatorial space of largely non-functional variants. Protein sequences, in particular, exemplify this structure, yet most generative artificial intelligence models implicitly assume a homogeneous data distribution. We show that this assumption fundamentally breaks down in heterogeneous biological sequence spaces: fixed global noise levels impose a destructive trade-off, either oversmoothing dense functional clusters or fragmenting sparse regions and producing non-functional hallucinations. To address this limitation, we introduce \emph{Density-Dependent Smoothing} (DDS), a geometry-aware generative framework that adapts stochastic smoothing to the local density of the underlying sequence landscape. By inversely coupling diffusion noise to estimated sequence density, DDS enables gentle refinement in high-density functional regions while promoting controlled exploration across sparse regions. Implemented as a plug-in mechanism for discrete molecular sampling, DDS consistently outperforms state-of-the-art diffusion and autoregressive models across antibody repertoires, therapeutic antibody design, antimicrobial peptide generation and coronavirus antibody design. Together, these results show that fixed global smoothing assumptions fundamentally limit generative modeling in sparse biological sequence spaces, and that geometry-aware smoothing removes this constraint, enabling reliable exploration and design previously unattainable with fixed-noise generative models.

cs.CE↗

arXiv2025

Knolling Bot: Teaching Robots the Human Notion of Tidiness

For robots to truly collaborate and assist humans, they must understand not only logic and instructions, but also the subtle emotions, aesthetics, and feelings that define our humanity. Human art and aesthetics are among the most elusive concepts-often difficult even for people to articulate-and without grasping these fundamentals, robots will be unable to help in many spheres of daily life. Consider the long-promised robotic butler: automating domestic chores demands more than motion planning. It requires an internal model of cleanliness and tidiness-a challenge largely unexplored by AI. To bridge this gap, we propose an approach that equips domestic robots to perform simple tidying tasks via knolling, the practice of arranging scattered items into neat, space-efficient layouts. Unlike the uniformity of industrial settings, household environments feature diverse objects and highly subjective notions of tidiness. Drawing inspiration from NLP, we treat knolling as a sequential prediction problem and employ a transformer based model to forecast each object's placement. Our method learns a generalizable concept of tidiness, generates diverse solutions adaptable to varying object sets, and incorporates human preferences for personalized arrangements. This work represents a step forward in building robots that internalize human aesthetic sense and can genuinely co-create in our living spaces.

cs.RO↗

arXiv2025

Uncertainty-Guided Model Selection for Tabular Foundation Models in Biomolecule Efficacy Prediction

In-context learners like TabPFN are promising for biomolecule efficacy prediction, where established molecular feature sets and relevant experimental results can serve as powerful contextual examples. However, their performance is highly sensitive to the provided context, making strategies like post-hoc ensembling of models trained on different data subsets a viable approach. An open question is how to select the best models for the ensemble without access to ground truth labels. In this study, we investigate an uncertainty-guided strategy for model selection. We demonstrate on an siRNA knockdown efficacy task that a TabPFN model using straightforward sequence-based features can surpass specialized state-of-the-art predictors. We also show that the model's predicted inter-quantile range (IQR), a measure of its uncertainty, has a negative correlation with true prediction error. We developed the OligoICP method, which selects and averages an ensemble of models with the lowest mean IQR for siRNA efficacy prediction, achieving superior performance compared to naive ensembling or using a single model trained on all available data. This finding highlights model uncertainty as a powerful, label-free heuristic for optimizing biomolecule efficacy predictions.

cs.LG↗

arXiv2025

Imitation Learning for Satellite Attitude Control under Unknown Perturbations

This paper presents a novel satellite attitude control framework that integrates Soft Actor-Critic (SAC) reinforcement learning with Generative Adversarial Imitation Learning (GAIL) to achieve robust performance under various unknown perturbations. Traditional control techniques often rely on precise system models and are sensitive to parameter uncertainties and external perturbations. To overcome these limitations, we first develop a SAC-based expert controller that demonstrates improved resilience against actuator failures, sensor noise, and attitude misalignments, outperforming our previous results in several challenging scenarios. We then use GAIL to train a learner policy that imitates the expert's trajectories, thereby reducing training costs and improving generalization through expert demonstrations. Preliminary experiments under single and combined perturbations show that the SAC expert can rotate the antenna to a specified direction and keep the antenna orientation reliably stable in most of the listed perturbations. Additionally, the GAIL learner can imitate most of the features from the trajectories generated by the SAC expert. Comparative evaluations and ablation studies confirm the effectiveness of the SAC algorithm and reward shaping. The integration of GAIL further reduces sample complexity and demonstrates promising imitation capabilities, paving the way for more intelligent and autonomous spacecraft control systems.

eess.SY↗

arXiv2024

Mixed Finite Element Method for Multi-layer Elastic Contact Systems

With the development of multi-layer elastic systems in the field of engineering mechanics, the corresponding variational inequality theory and algorithm design have received more attention and research. In this study, a class of equivalent saddle point problems with interlayer Tresca friction conditions and the mixed finite element method are proposed and analyzed. Then, the convergence of the numerical solution of the mixed finite element method is theoretically proven, and the corresponding algebraic dual algorithm is given. Finally, through numerical experiments, the mixed finite element method is not only compared with the layer decomposition method, but also its convergence relationship with respect to the spatial discretization parameter $H$ is verified.

math.NA↗

arXiv2024

A layer decomposition method for multi-layer elastic contact systems with interlayer Tresca friction

With the increasing demand for the accuracy of numerical simulation of pavement mechanics, the variational inequality model and its induced finite element method which can simulate the interlayer contact state becomes a potential solution. In this paper, a layer decomposition algorithm for solving variational inequality models of multi-layer elastic contact systems with interlayer Tresca friction conditions is studied. Continuous and discrete versions of the algorithm and their convergence theorems have been proposed and proved successively. Then, the algebraic form of the executable optimization algorithm and the numerical experimental results verify the practicability of the variational inequality model and its algorithm in the pavement mechanics modeling.

math.NA↗

arXiv2024

Variational inequalities of multilayer elastic systems with interlayer friction: existence and uniqueness of solution and convergence of numerical solution

Based on the mathematical-physical model of pavement mechanics, a multilayer elastic system with interlayer friction conditions is constructed. Given the complex boundary conditions, the corresponding variational inequalities of the partial differential equations are derived, so that the problem can be analyzed under the variational framework. First, the existence and uniqueness of the solution of the variational inequality is proved; then the approximation error of the numerical solution based on the finite element method is analyzed, and when the finite element space satisfies certain approximation conditions, the convergence of the numerical solution is proved; finally, in the trivial finite element space, the convergence order of the numerical solution is derived. The above conclusions provide basic theoretical support for solving the displacement-strain problem of multilayer elastic systems under the framework of variational inequalities.

math.NA↗

arXiv2023

Knolling bot 2.0: Enhancing Object Organization with Self-supervised Graspability Estimation

Building on recent advancements in transformer based approaches for domestic robots performing knolling, the art of organizing scattered items into neat arrangements. This paper introduces Knolling bot 2.0. Recognizing the challenges posed by piles of objects or items situated closely together, this upgraded system incorporates a self-supervised graspability estimation model. If objects are deemed ungraspable, an additional behavior will be executed to separate the objects before knolling the table. By integrating this grasp prediction mechanism with existing visual perception and transformer based knolling models, an advanced system capable of decluttering and organizing even more complex and densely populated table settings is demonstrated. Experimental evaluations demonstrate the effectiveness of this module, yielding a graspability prediction accuracy of 95.7%.

cs.RO↗

arXiv2022

The Mean Time to Absorption on Horizontal Partitioned Sierpinski Gasket Networks

The random walk is one of the most basic dynamic properties of complex networks, which has gradually become a research hotspot in recent years due to its many applications in actual networks. An important characteristic of the random walk is the mean time to absorption, which plays an extremely important role in the study of topology, dynamics and practical application of complex networks. Analyzing the mean time to absorption on the regular iterative self-similar network models is an important way to explore the influence of self-similarity on the properties of random walks on the network. The existing literatures have proved that even local self-similar structures can greatly affect the properties of random walks on the global network, but they have failed to prove whether these effects are related to the scale of these self-similar structures. In this article, we construct and study a class of Horizontal Partitioned Sierpinski Gasket network model based on the classic Sierpinski gasket network, which is composed of local self-similar structures, and the scale of these structures will be controlled by the partition coefficient $k$. Then, the analytical expressions and approximate expressions of the mean time to absorption on the network model are obtained, which prove that the size of the self-similar structure in the network will directly restrict the influence of the self-similar structure on the properties of random walks on the network. Finally, we also analyzed the mean time to absorption of different absorption nodes on the network to find the location of the node with the highest absorption efficiency.

physics.soc-ph↗

arXiv2022

Spectral Analysis and its applications for a class of scale-free network based on the weighted m-clique annex operation

The spectrum of network is an important tool to study the function and dynamic properties of network, and graph operation and product is an effective mechanism to construct a specific local and global topological structure. In this study, a class of weighted $m-$clique annex operation $τ_m^r(\cdot)$ controlled by scale factor $m$ and weight factor $r$ is defined, through which an iterative weighted network model $G_t$ with small-world and scale-free properties is constructed. In particular, when the number of iterations $t$ tends to infinity, the network has transfinite fractal property. Then, through the iterative features of the network structure, the iterative relationship of the eigenvalues of the normalized Laplacian matrix corresponding to the network is studied. Accordingly, some applications of the spectrum of the network, including the Kenemy constant, Multiplicative Degree-Kirchhoff index and the number of weighted spanning trees, are further given. In addition, we also study the effect of the two factors controlling network operation on the structure and function of the iterative weighted network $G_t$, so that the network operation can better simulate the real network and have more application potential in the field of artificial network.

math.DS↗

arXiv2021

Average trapping time on a type of horizontally segmented 3 dimensional Sierpinski gasket network with two types of locally self-similar structures

As a classic self-similar network model, Sierpinski gasket network has been used many times to study the characteristics of self-similar structure and its influence on the dynamic properties of the network. However, the network models studied in these problems only contain a single self-similar structure, which is inconsistent with the structural characteristics of the actual network models. In this paper, a type of horizontally segmented 3 dimensional Sierpinski gasket network is constructed, whose main feature is that it contains the locally self-similar structures of the 2 dimensional Sierpinski gasket network and the 3 dimensional Sierpinski gasket network at the same time, and the scale transformation between the two kinds of self-similar structures can be controlled by adjusting the crosscutting coefficient. The analytical expression of the average trapping time on the network model is solved, which used to analyze the effect of two types of self-similar structures on the properties of random walks. Finally, we conclude that the dominant self-similar structure will exert a greater influence on the random walk process on the network.

cond-mat.stat-mech↗

arXiv2021

Average Trapping time on the 3-dimensional 3-level Sierpinski gasket network with a set of trap nodes

As a basic dynamic feature on complex networks, the property of random walk has received a lot of attention in recent years. In this paper, we first studied the analytical expression of the mean global first passage time (MGFPT) on the 3-dimensional 3-level Sierpinski gasket network. Based on the self-similar structure of the network, the correlation between the MGFPT and the average trapping time (ATT) is found, and then the analytical expression of the ATT is obtained. Finally, by establishing a joint network model, we further give the standard process of solving the analytical expression of the ATT when there is a set of trap nodes in the network. By illustrating examples and numerical simulations, it can be proved that when the trap node sets are different, the ATT will be quite different, but the the super-linear relationship with the number of iterations will not be changed.

cond-mat.stat-mech↗

arXiv2020

Genie: A Secure, Transparent Sharing and Services Platform for Genetic and Health Data

Artificial Intelligence (AI) incorporating genetic and medical information have been applied in disease risk prediction, unveiling disease mechanism, and advancing therapeutics. However, AI training relies on highly sensitive and private data which significantly limit their applications and robustness evaluation. Moreover, the data access management after sharing across organization heavily relies on legal restriction, and there is no guarantee in preventing data leaking after sharing. Here, we present Genie, a secure AI platform which allows AI models to be trained on medical data securely. The platform combines the security of Intel Software Guarded eXtensions (SGX), transparency of blockchain technology, and verifiability of open algorithms and source codes. Genie shares insights of genetic and medical data without exposing anyone's raw data. All data is instantly encrypted upon upload and contributed to the models that the user chooses. The usage of the model and the value generated from the genetic and health data will be tracked via a blockchain, giving the data transparent and immutable ownership.

cs.CR↗