arXiv subjects

Yihao Zhang

Publications and source records attributed to Yihao Zhang.

59 records · Page 4Linked to original sources

arXiv2020

A deep reinforcement learning model based on deterministic policy gradient for collective neural crest cell migration

Modeling cell interactions such as co-attraction and contact-inhibition of locomotion is essential for understanding collective cell migration. Here, we propose a novel deep reinforcement learning model for collective neural crest cell migration. We apply the deep deterministic policy gradient algorithm in association with a particle dynamics simulation environment to train agents to determine the migration path. Because of the different migration mechanisms of leader and follower neural crest cells, we train two types of agents (leaders and followers) to learn the collective cell migration behavior. For a leader agent, we consider a linear combination of a global task, resulting in the shortest path to the target source, and a local task, resulting in a coordinated motion along the local chemoattractant gradient. For a follower agent, we consider only the local task. First, we show that the self-driven forces learned by the leader cell point approximately to the placode, which means that the agent is able to learn to follow the shortest path to the target. To validate our method, we compare the total time elapsed for agents to reach the placode computed using the proposed method and the time computed using an agent-based model. The distributions of the migration time intervals calculated using the two methods are shown to not differ significantly. We then study the effect of co-attraction and contact-inhibition of locomotion to the collective leader cell migration. We show that the overall leader cell migration for the case with co-attraction is slower because the co-attraction mitigates the source-driven effect. In addition, we find that the leader and follower agents learn to follow a similar migration behavior as in experimental observations. Overall, our proposed method provides useful insight on how to apply reinforcement learning techniques to simulate collective cell migration.

q-bio.CB↗

arXiv2020

Scheduling with Communication Delays via LP Hierarchies and Clustering

We consider the classic problem of scheduling jobs with precedence constraints on identical machines to minimize makespan, in the presence of communication delays. In this setting, denoted by $\mathsf{P} \mid \mathsf{prec}, c \mid C_{\mathsf{max}}$, if two dependent jobs are scheduled on different machines, then at least $c$ units of time must pass between their executions. Despite its relevance to many applications, this model remains one of the most poorly understood in scheduling theory. Even for a special case where an unlimited number of machines is available, the best known approximation ratio is $2/3 \cdot (c+1)$, whereas Graham's greedy list scheduling algorithm already gives a $(c+1)$-approximation in that setting. An outstanding open problem in the top-10 list by Schuurman and Woeginger and its recent update by Bansal asks whether there exists a constant-factor approximation algorithm. In this work we give a polynomial-time $O(\log c \cdot \log m)$-approximation algorithm for this problem, where $m$ is the number of machines and $c$ is the communication delay. Our approach is based on a Sherali-Adams lift of a linear programming relaxation and a randomized clustering of the semimetric space induced by this lift.

cs.DS↗

arXiv2018

Local Contrast Learning

Learning a deep model from small data is yet an opening and challenging problem. We focus on one-shot classification by deep learning approach based on a small quantity of training samples. We proposed a novel deep learning approach named Local Contrast Learning (LCL) based on the key insight about a human cognitive behavior that human recognizes the objects in a specific context by contrasting the objects in the context or in her/his memory. LCL is used to train a deep model that can contrast the recognizing sample with a couple of contrastive samples randomly drawn and shuffled. On one-shot classification task on Omniglot, the deep model based LCL with 122 layers and 1.94 millions of parameters, which was trained on a tiny dataset with only 60 classes and 20 samples per class, achieved the accuracy 97.99% that outperforms human and state-of-the-art established by Bayesian Program Learning (BPL) trained on 964 classes. LCL is a fundamental idea which can be applied to alleviate parametric model's overfitting resulted by lack of training samples.

cs.LG↗

arXiv2016

Axon Membrane Skeleton Structure is Optimized for Coordinated Sodium Propagation

Axons transmit action potentials with high fidelity and minimal jitter. This unique capability is likely the result of the spatiotemporal arrangement of sodium channels along the axon. Super-resolution microscopy recently revealed that the axon membrane skeleton is structured as a series of actin rings connected by spectrin filaments that are held under entropic tension. Sodium channels also exhibit a periodic distribution pattern, as they bind to ankyrin G, which associates with spectrin. Here, we elucidate the relationship between the axon membrane skeleton structure and the function of the axon. By combining cytoskeletal dynamics and continuum diffusion modeling, we show that spectrin filaments under tension minimize the thermal fluctuations of sodium channels and prevent overlap of neighboring channel trajectories. Importantly, this axon skeletal arrangement allows for a highly reproducible band-like activation of sodium channels leading to coordinated sodium propagation along the axon.

physics.bio-ph↗

arXiv2015

Facial Expression Cloning with Elastic and Muscle Models

Expression cloning plays an important role in facial expression synthesis. In this paper, a novel algorithm is proposed for facial expression cloning. The proposed algorithm first introduces a new elastic model to balance the global and local warping effects, such that the impacts from facial feature diversity among people can be minimized, and thus more effective geometric warping results can be achieved. Furthermore, a muscle-distribution-based (MD) model is proposed, which utilizes the muscle distribution of the human face and results in more accurate facial illumination details. In addition, we also propose a new distance-based metric to automatically select the optimal parameters such that the global and local warping effects in the elastic model can be suitably balanced. Experimental results show that our proposed algorithm outperforms the existing methods.

cs.GR↗