arXiv subjects

Chengjun Wu

Publications and source records attributed to Chengjun Wu.

6 recordsLinked to original sources

arXiv2024

BioRAG: A RAG-LLM Framework for Biological Question Reasoning

The question-answering system for Life science research, which is characterized by the rapid pace of discovery, evolving insights, and complex interactions among knowledge entities, presents unique challenges in maintaining a comprehensive knowledge warehouse and accurate information retrieval. To address these issues, we introduce BioRAG, a novel Retrieval-Augmented Generation (RAG) with the Large Language Models (LLMs) framework. Our approach starts with parsing, indexing, and segmenting an extensive collection of 22 million scientific papers as the basic knowledge, followed by training a specialized embedding model tailored to this domain. Additionally, we enhance the vector retrieval process by incorporating a domain-specific knowledge hierarchy, which aids in modeling the intricate interrelationships among each query and context. For queries requiring the most current information, BioRAG deconstructs the question and employs an iterative retrieval process incorporated with the search engine for step-by-step reasoning. Rigorous experiments have demonstrated that our model outperforms fine-tuned LLM, LLM with search engines, and other scientific RAG frameworks across multiple life science question-answering tasks.

cs.CL↗

arXiv2024

Pretrained Hybrids with MAD Skills

While Transformers underpin modern large language models (LMs), there is a growing list of alternative architectures with new capabilities, promises, and tradeoffs. This makes choosing the right LM architecture challenging. Recently proposed hybrid architectures seek a best-of-all-worlds approach that reaps the benefits of all architectures. Hybrid design is difficult for two reasons: it requires manual expert-driven search, and new hybrids must be trained from scratch. We propose Manticore, a framework that addresses these challenges by automating the design of hybrid architectures while reusing pretrained models to create pretrained hybrids. Our approach augments ideas from differentiable Neural Architecture Search (NAS) by incorporating simple projectors that translate features between pretrained blocks from different architectures. We then fine-tune hybrids that combine pretrained models from different architecture families -- such as the GPT series and Mamba -- end-to-end. With Manticore, we enable LM selection without training multiple models, the construction of pretrained hybrids from existing pretrained models, and the ability to program pretrained hybrids to have certain capabilities. Manticore hybrids match existing manually designed hybrids, achieve strong performance on Long Range Arena, and improve on pretrained transformers and state space models on various natural language tasks.

cs.LG↗

arXiv2014

Quantum resource studied from the perspective of quantum state superposition

Quantum resources,such as discord and entanglement, are crucial in quantum information processing. In this paper, quantum resources are studied from the aspect of quantum state superposition. We define the local superposition (LS) as the superposition between basis of single part, and nonlocal superposition (NLS) as the superposition between product basis of multiple parts. For quantum resource with nonzero LS, quantum operation must be introduced to prepare it, and for quantum resource with nonzero NLS, nonlocal quantum operation must be introduced to prepare it. We prove that LS vanishes if and only if the state is classical and NLS vanishes if and only if the state is separable. From this superposition aspect, quantum resources are categorized as superpositions existing in different parts. These results are helpful to study quantum resources from a unified frame.

quant-ph↗

arXiv2014

Quantum Evolution Time Limit: a General Non-Markovian Bound

We derive a sharp bound as the quantum speed limit (QSL) for the minimal evolution time of quantum open systems in the non-Markovian strong-coupling regime with initial mixed states by considering the effects of both renormalized Hamiltonian and dissipator. For a non-Markovian quantum open system, the possible evolution time between two arbitrary states is not unique, among the set of which we find that the minimal one and its QSL can decrease more steeply by adjusting the coupling strength of the dissipator, which thus provides potential improvements of efficiency in many quantum physics and quantum information areas.

quant-ph↗

arXiv2011

Non-Markovian dynamics without using quantum trajectory

Open quantum system interacting with structured environment is important and manifests non- Markovian behavior, which was conventionally studied using quantum trajectory stochastic method. In this paper, by dividing the effects of the environment into two parts, we propose a deterministic method without using quantum trajectory. This method is more efficient and accurate than stochastic method in most Markovian and non-Markovian cases. We also extend this method to the generalized Lindblad master equation.

quant-ph↗

arXiv2009

Non-Markovian Dynamics of Entanglement for Multipartite Systems

Entanglement dynamics for a couple of two-level atoms interacting with independent structured reservoirs is studied using a non-perturbative approach. It is shown that the revival of atom entanglement is not necessarily accompanied by the sudden death of reservoir entanglement, and vice versa. In fact, atom entanglement can revive before, simultaneously or even after the disentanglement of reservoirs. Using a novel method based on the population analysis for the excited atomic state, we present the quantitative criteria for the revival and death phenomena. For giving a more physically intuitive insight, the quasimode Hamiltonian method is applied. Our quantitative analysis is helpful for the practical engineering of entanglement.

quant-ph↗