arXiv subjects

Jiacheng Fan

Publications and source records attributed to Jiacheng Fan.

5 recordsLinked to original sources

arXiv2026

RoboPaint: From Human Demonstration to Any Robot and Any View

Acquiring large-scale, high-fidelity robot demonstration data remains a critical bottleneck for scaling Vision-Language-Action (VLA) models in dexterous manipulation. We propose a Real-Sim-Real data collection and data editing pipeline that transforms human demonstrations into robot-executable, environment-specific training data without direct robot teleoperation. Standardized data collection rooms are built to capture multimodal human demonstrations (synchronized 3 RGB-D videos, 11 RGB videos, 29-DoF glove joint angles, and 14-channel tactile signals). Based on these human demonstrations, we introduce a tactile-aware retargeting method that maps human hand states to robot dex-hand states via geometry and force-guided optimization. Then the retargeted robot trajectories are rendered in a photorealistic Isaac Sim environment to build robot training data. Real world experiments have demonstrated: (1) The retargeted dex-hand trajectories achieve an 84\% success rate across 10 diverse object manipulation tasks. (2) VLA policies (Pi0.5) trained exclusively on our generated data achieve 80\% average success rate on three representative tasks, i.e., pick-and-place, pushing and pouring. To conclude, robot training data can be efficiently "painted" from human demonstrations using our real-sim-real data pipeline. We offer a scalable, cost-effective alternative to teleoperation with minimal performance loss for complex dexterous manipulation.

cs.RO↗

arXiv2025

Dynamic Asset Pricing with {\alpha}-MEU Model

We study a dynamic asset pricing problem in which a representative agent is ambiguous about the aggregate endowment growth rate and trades a risky stock, human capital, and a risk-free asset to maximize her preference value of consumption represented by the {\alpha}-maxmin expected utility model. This preference model is known to be dynamically inconsistent, so we consider intra-personal equilibrium strategies for the representative agent and define the market equilibrium as the one in which the strategy that clears the market is an intra-personal equilibrium. We prove the existence and uniqueness of the market equilibrium and show that the asset prices in the equilibrium are the same as in the case when the agent does not perceive any ambiguity but believes in a particular probabilistic model of the endowment process. We show that with reasonable parameter values, the more ambiguity the agent perceives or the more ambiguity-averse she is, the lower the risk-free rate, the higher the stock price, the higher the stock risk premium, and the lower the stock volatility.

q-fin.PR↗

arXiv2025

Quantum Knowledge Distillation for Large Language Models

As foundational tools in natural language processing, Large Language Models (LLMs) have immense parameter scales, which makes deployment and inference increasingly prohibitive, especially in resource-constrained devices. Therefore, knowledge distillation for LLMs, i.e., compressing the LLM to a smaller model, is meaningful. With strong parameter representation capacity, quantum computing is regarded as a promising solution. Here, we propose a Quantum knowledge Distillation model for LLMs (QD-LLM) that leverages variational quantum circuits to learn from LLMs. In classical simulation, QD-LLM outperforms several mainstream distillation methods on multiple text classification tasks in terms of both accuracy and efficiency using only 11 qubits. The results reveal an interesting phenomenon that the simulation of quantum student models may be regarded as a new class of quantum-inspired classical algorithms. Remarkably, we deploy the obtained circuits on the Baihua superconducting quantum processor via the Quafu platform to assess practical feasibility. The model maintains stable inference performance despite hardware constraints such as decoherence and finite sampling. In summary, QD-LLM marks a foundational step in connecting quantum computing with LLMs, demonstrating the feasibility of quantum-native approaches that aim to compress and deploy models of increasingly larger scales. The code of this article has been open-sourced at https://github.com/Lilingxiao-bupt/QD-LLM.

quant-ph↗

arXiv2025

Topology-Driven Quantum Architecture Search Framework

The limitations of Noisy Intermediate-Scale Quantum (NISQ) devices have motivated the development of Variational Quantum Algorithms (VQAs), which are designed to potentially achieve quantum advantage for specific tasks. Quantum Architecture Search (QAS) algorithms play a critical role in automating the design of high-performance Parameterized Quantum Circuits (PQCs) for VQAs. However, existing QAS approaches struggle with large search spaces, leading to substantial computational overhead when optimizing large-scale quantum circuits. Extensive empirical analysis reveals that circuit topology has a greater impact on quantum circuit performance than gate types. Based on this insight, we propose the Topology-Driven Quantum Architecture Search (TD-QAS) framework, which first identifies optimal circuit topologies and then fine-tunes the gate types. In the fine-tuning phase, the QAS inherits parameters from the topology search phase, eliminating the need for training from scratch. By decoupling the large search space into separate topology and gate-type components, TD-QAS avoids exploring gate configurations within low-performance topologies, thereby significantly reducing computational complexity. Numerical simulations across various tasks, under both noiseless and noisy conditions, validate the effectiveness of the TD-QAS framework. This framework advances standard QAS algorithms by enabling the identification of high-performance quantum circuits while minimizing computational demands. These findings indicate that TD-QAS deepens our understanding of VQAs and offers broad potential for the development of future QAS algorithms.

quant-ph↗

arXiv2023

Optimal dividend payout with path-dependent drawdown constraint

This paper studies an optimal dividend problem with a drawdown constraint in a Brownian motion model, requiring the dividend payout rate to remain above a fixed proportion of its historical maximum. This leads to a path-dependent stochastic control problem, as the admissible control depends on its own past values. The associated Hamilton-Jacobi-Bellman (HJB) equation is a novel two-dimensional variational inequality with a gradient constraint, a type of problem previously only analyzed in the literature using viscosity solution techniques. In contrast, this paper employs delicate PDE methods to establish the existence of a strong solution. This stronger regularity allows us to explicitly characterize an optimal feedback control strategy, expressed in terms of two free boundaries and the running maximum surplus process. Furthermore, we derive key properties of the value function and the free boundaries, including boundedness and continuity. Numerical examples are provided to verify the theoretical results and to offer new financial insights.

q-fin.MF↗