arXiv subjects

Zhen Luo

Publications and source records attributed to Zhen Luo.

25 records · Page 2Linked to original sources

arXiv2024

LLplace: The 3D Indoor Scene Layout Generation and Editing via Large Language Model

Designing 3D indoor layouts is a crucial task with significant applications in virtual reality, interior design, and automated space planning. Existing methods for 3D layout design either rely on diffusion models, which utilize spatial relationship priors, or heavily leverage the inferential capabilities of proprietary Large Language Models (LLMs), which require extensive prompt engineering and in-context exemplars via black-box trials. These methods often face limitations in generalization and dynamic scene editing. In this paper, we introduce LLplace, a novel 3D indoor scene layout designer based on lightweight fine-tuned open-source LLM Llama3. LLplace circumvents the need for spatial relationship priors and in-context exemplars, enabling efficient and credible room layout generation based solely on user inputs specifying the room type and desired objects. We curated a new dialogue dataset based on the 3D-Front dataset, expanding the original data volume and incorporating dialogue data for adding and removing objects. This dataset can enhance the LLM's spatial understanding. Furthermore, through dialogue, LLplace activates the LLM's capability to understand 3D layouts and perform dynamic scene editing, enabling the addition and removal of objects. Our approach demonstrates that LLplace can effectively generate and edit 3D indoor layouts interactively and outperform existing methods in delivering high-quality 3D design solutions. Code and dataset will be released.

cs.CV↗

arXiv2024

Local existence of classical solution to the chemotaxis-shallow water system with vacuum in $\mathbb{R}^2$

In this paper, we consider the chemotaxis-shallow water system in $\mathbb{R}^2$. We establish the local existence of classical solution without assuming the initial height is small or has a small perturbation near a constant. The far field behavior of the height is a constant which could be either vacuum or non-vacuum. The initial data is allowed vacuum and the spatial measure of the set of vacuum can be arbitrarily large.

math.AP↗

arXiv2023

YAYI 2: Multilingual Open-Source Large Language Models

As the latest advancements in natural language processing, large language models (LLMs) have achieved human-level language understanding and generation abilities in many real-world tasks, and even have been regarded as a potential path to the artificial general intelligence. To better facilitate research on LLMs, many open-source LLMs, such as Llama 2 and Falcon, have recently been proposed and gained comparable performances to proprietary models. However, these models are primarily designed for English scenarios and exhibit poor performances in Chinese contexts. In this technical report, we propose YAYI 2, including both base and chat models, with 30 billion parameters. YAYI 2 is pre-trained from scratch on a multilingual corpus which contains 2.65 trillion tokens filtered by our pre-training data processing pipeline. The base model is aligned with human values through supervised fine-tuning with millions of instructions and reinforcement learning from human feedback. Extensive experiments on multiple benchmarks, such as MMLU and CMMLU, consistently demonstrate that the proposed YAYI 2 outperforms other similar sized open-source models.

cs.CL↗

arXiv2023

PCPNet: An Efficient and Semantic-Enhanced Transformer Network for Point Cloud Prediction

The ability to predict future structure features of environments based on past perception information is extremely needed by autonomous vehicles, which helps to make the following decision-making and path planning more reasonable. Recently, point cloud prediction (PCP) is utilized to predict and describe future environmental structures by the point cloud form. In this letter, we propose a novel efficient Transformer-based network to predict the future LiDAR point clouds exploiting the past point cloud sequences. We also design a semantic auxiliary training strategy to make the predicted LiDAR point cloud sequence semantically similar to the ground truth and thus improves the significance of the deployment for more tasks in real-vehicle applications. Our approach is completely self-supervised, which means it does not require any manual labeling and has a solid generalization ability toward different environments. The experimental results show that our method outperforms the state-of-the-art PCP methods on the prediction results and semantic similarity, and has a good real-time performance. Our open-source code and pre-trained models are available at https://github.com/Blurryface0814/PCPNet.

cs.CV↗

arXiv2018

Topology optimization of micro-structured materials featured with the specific mechanical properties

Micro-structured materials consisting of an array of microstructures are engineered to provide the specific material properties. This present work investigates the design of cellular materials under the framework of level set, so as to optimize the topologies and shapes of these porous material microstructures. Firstly, the energy-based homogenization method (EBHM) is applied to evaluate the material effective properties based on the topology of the material cell, where the effective elasticity property is evaluated by the average stress and strain theorems. Secondly, a parametric level set method (PLSM) is employed to optimize the microstructural topology until the specific mechanical properties can be achieved, including the maximum bulk modulus, the maximum shear modulus and their combinations, as well as the negative Poisson's ratio (NPR). The complicated topological shape optimization of the material microstructure has been equivalent to evolve the sizes of the expansion coefficients in the interpolation of the level set function. Finally, several numerical examples are fully discussed to demonstrate the effectiveness of the developed method. A series of new and interesting material cells with the specific mechanical properties can be found.

cs.CE↗

arXiv2018

Externally-Contracted Multi-Reference Configuration Interaction Method Using a DMRG Reference Wave Function

The recent development of the density matrix renormalization group (DMRG) method in multireference quantum chemistry makes it practical to evaluate static correlation in a large active space, while dynamic correlation provides a critical correction to the DMRG reference for strong-correlated systems and is usually obtained using multi-reference perturbation (MRPT) or configuration interaction (MRCI) methods with internal contraction (ic) approximation. These methods can use active space scalable to relatively larger size references than has previously been possible. However, they are still hardly applicable to systems with active space larger than 30 orbitals because of high computation and storage costs of high-order reduced density matrices (RDMs) and the number of virtual orbitals are normally limited to few hundreds. In this work, we propose a new effective implementation of DMRG-MRCI, in which we use re-constructed CASCI-type configurations from DMRG wave function via the entropy-driving genetic algorithm (EDGA), and integrate with MRCI by an external contraction (ec) scheme. This bypasses the bottleneck of computing high-order RDMs in traditional DMRG dynamic correlation methods with ic approximation and the number of MRCI configurations is not dependent on the number of virtual orbitals. Therefore, DMRG-ec-MRCI method is promising for dealing with larger active space than 30 orbitals and large basis sets. We demonstrate the capability of our DMRG-ec-MRCI method in several benchmark applications, including the evaluation of potential energy curve of Cr$_{2}$, single-triplet gaps of higher n-acene molecules and the energy of Eu-BTBP(NO$_3$)$_3$ complex.

cond-mat.str-el↗

arXiv2017

Efficient reconstruction of CASCI-type wave functions for a DMRG state using quantum information theory and genetic algorithm

We improve the methodology to construct a complete active space-configuration interaction (CAS-CI) expansion for density-matrix renormalization group (DMRG) wave function using matrix-product state representation, inspired by the sampling-reconstructed CAS [SR-CAS, Boguslawski et al, J. Chem. Phys. 2011, 134, 224101] algorithm. In our scheme, a genetic algorithm, in which the "crossover" and "mutation" process can be optimized based on quantum information theory, is employed when reconstructing the CASCI-type wave function in the Hilbert space. Test analysis results for the ground and excited state wave functions of conjugated molecules and transition metal compounds illustrate that our scheme is very efficient for searching the most important CI expansions in large active spaces.

cond-mat.str-el↗