arXiv subjects

Junda Wu

Publications and source records attributed to Junda Wu.

75 records · Page 5Linked to original sources

arXiv2023

Computationally efficient human body modelling for real time motion comfort assessment

Due to the complexity of the human body and its neuromuscular stabilization, it has been challenging to efficiently and accurately predict human motion and capture posture while being driven. Existing simple models of the seated human body are mostly two-dimensional and developed in the mid-sagittal plane ex-posed to in-plane excitation. Such models capture fore-aft and vertical motion but not the more complex 3D motions due to lateral loading. Advanced 3D full-body active human models (AHMs), such as in MADYMO, can be used for comfort analysis and to investigate how vibrations influence the human body while being driven. However, such AHMs are very time-consuming due to their complexity. To effectively analyze motion comfort, a computationally efficient and accurate three dimensional (3D) human model, which runs faster than real-time, is presented. The model's postural stabilization parameters are tuned using available 3D vibration data for head, trunk and pelvis translation and rotation. A comparison between AHM and EHM is conducted regarding human body kinematics. According to the results, the EHM model configuration with two neck joints, two torso bending joints, and a spinal compression joint accurately predicts body kinematics.

cs.HC↗

arXiv2023

InfoPrompt: Information-Theoretic Soft Prompt Tuning for Natural Language Understanding

Soft prompt tuning achieves superior performances across a wide range of few-shot tasks. However, the performances of prompt tuning can be highly sensitive to the initialization of the prompts. We also empirically observe that conventional prompt tuning methods cannot encode and learn sufficient task-relevant information from prompt tokens. In this work, we develop an information-theoretic framework that formulates soft prompt tuning as maximizing mutual information between prompts and other model parameters (or encoded representations). This novel view helps us to develop a more efficient, accurate and robust soft prompt tuning method InfoPrompt. With this framework, we develop two novel mutual information based loss functions, to (i) discover proper prompt initialization for the downstream tasks and learn sufficient task-relevant information from prompt tokens and (ii) encourage the output representation from the pretrained language model to be more aware of the task-relevant information captured in the learnt prompt. Extensive experiments validate that InfoPrompt can significantly accelerate the convergence of the prompt tuning and outperform traditional prompt tuning methods. Finally, we provide a formal theoretical result for showing to show that gradient descent type algorithm can be used to train our mutual information loss.

cs.CL↗

arXiv2023

Characterization, synthesis, and optimization of quantum circuits over multiple-control $\textit{Z}$-rotation gates: A systematic study

We conduct a systematic study of quantum circuits composed of multiple-control $Z$-rotation (MCZR) gates as primitives, since they are widely-used components in quantum algorithms and also have attracted much experimental interest in recent years. Herein, we establish a circuit-polynomial correspondence to characterize the functionality of quantum circuits over the MCZR gate set with continuous parameters. An analytic method for exactly synthesizing such quantum circuit to implement any given diagonal unitary matrix with an optimal gate count is proposed, which also enables the circuit depth optimal for specific cases with pairs of complementary gates. Furthermore, we present a gate-exchange strategy together with a flexible iterative algorithm for effectively optimizing the depth of any MCZR circuit, which can also be applied to quantum circuits over any other commuting gate set. Besides the theoretical analysis, the practical performances of our circuit synthesis and optimization techniques are further evaluated by numerical experiments on two typical examples in quantum computing, including diagonal Hermitian operators and Quantum Approximate Optimization Algorithm (QAOA) circuits with tens of qubits, which can demonstrate a reduction in circuit depth by 33.40\% and 15.55\% on average over relevant prior works, respectively. Therefore, our methods and results provide a pathway for implementing quantum circuits and algorithms on recently developed devices.

quant-ph↗