Jidong Tian
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jidong Tian has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Hao He
State Key Laboratory of Photonics and Communications, Shanghai Jiao Tong University, Shanghai, China
2 shared publications - Caoyun Fan
MoE Key Lab of Artificial Intelligence AI Institute Shanghai Jiao Tong University Shanghai China
1 shared publication - Chao Wang
School of Aerospace Engineering, Huazhong University of Science and Technology 2 , 430074 Wuhan
1 shared publication - Hongxing Hua
State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai, China
1 shared publication - Min Tang
Shanghai Jiao Tong University
1 shared publication - Shanxing Mei
State Key Lab of Advanced Optical Communication System and Network, Shanghai Jiao Tong University, China
1 shared publication
Latest Publications
- Open Access
- 8 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 546-552
LDSeq: Latent Diffusion Models for Sequence to Sequence Text Generation
Diffusion models have demonstrated remarkable success in generating continuous data, such as images and audios. Previous studies on text generation employing continuous diffusion models have revealed the potential of the diffusion framework. However, challenges like embedding collapse persist, limiting the overall generation performance. In this paper we introduce LDSeq, a latent diffusion framework employing a two-stage training procedure for sequence-to-sequence text generation. In the proposed framework, we first train a Variational Auto-Encoder (VAE) on downstream datasets to compress the target text of samples into a continuous latent space, and then we train a conditional latent diffusion model in the fixed continuous latent space, where the latent vectors are iteratively sampled conditioned on the input source text. The disjoint training stages prevent the collapse of diffusion space. Experimental results on paraphrase generation and text summarization datasets show that LDSeq achieves comparable or superior performance in comparison to AR and NAR baselines while requiring lower training cost. Furthermore, We discuss some potential future directions for enhancing diffusion models in the text generation domain.
- 8 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 546-552
Open AccessLDSeq: Latent Diffusion Models for Sequence to Sequence Text Generation
Diffusion models have demonstrated remarkable success in generating continuous data, such as images and audios. Previous studies on text generation employing continuous diffusion models have revealed the potential of the diffusion framework. However, challenges like embedding collapse persist, limiting the overall generation performance. In this paper we introduce LDSeq, a latent diffusion framework employing a two-stage training procedure for sequence-to-sequence text generation. In the proposed framework, we first train a Variational Auto-Encoder (VAE) on downstream datasets to compress the target text of samples into a continuous latent space, and then we train a conditional latent diffusion model in the fixed continuous latent space, where the latent vectors are iteratively sampled conditioned on the input source text. The disjoint training stages prevent the collapse of diffusion space. Experimental results on paraphrase generation and text summarization datasets show that LDSeq achieves comparable or superior performance in comparison to AR and NAR baselines while requiring lower training cost. Furthermore, We discuss some potential future directions for enhancing diffusion models in the text generation domain.
- 11 October 2023
- Published by Elsevier in Expert Systems with Applications
- Vol. 238
Unlock the Potential of Counterfactually-Augmented Data in Out-Of-Distribution Generalization
- 11 October 2023
- Published by Elsevier in Expert Systems with Applications
- Vol. 238
Unlock the Potential of Counterfactually-Augmented Data in Out-Of-Distribution Generalization
- 8 February 2022
- Published by IOP Publishing in Smart Materials and Structures
- Vol. 31 (3) , 035023
Numerical and experimental analysis on the helicopter rotor dynamic load controlled by the actively trailing edge flap
Reducing the rotor dynamic load is an important issue to improve the performance and reliability of a helicopter. The control mechanism of the actively controlled flap (ACF) on the rotor dynamic load is numerically and experimentally investigated by a 3-blade helicopter rotor in this paper. In the aero-elastic numerical approach, the complex motion of the rotor such as the stretching, bending, torsion and pitching of the blade including the deflection of the ACF are all taken into consideration in the structural formulation. The aerodynamic solution adopted the vortex lattice method combined with the free wake model, in which the influence of ACF on the free wake and the aerodynamic load on the blade is taken into account as well. While the experimental method of measuring hub loads and acoustic was accomplished by a rotor rig in a wind tunnel. The result shows that the 3/rev ACF actuation can reduce the 3 ω hub load by more than 50% at maximum, which is significantly better than the 4/rev control. While 4/rev has greater potential to reduce blade vortex interaction (BVI) loads than 3/rev with µ = 0.15. Further mechanistic analysis shows that by changing the phase difference between the dynamic load on the flap and the rest of the blade, the peak load on the whole blade can be improved, thus achieving effective control of the hub dynamic load, the flap reaches the minimum angle of attack at 90 ∘ –100 ∘ azimuth under best control condition; when the BVI load is perfectly controlled, the flap reaches the minimum angle of attack at 140 ∘ azimuth, and by changing the circulation of the wake, the intensity of BVI in the advancing side is improved. Moreover, an interesting finding in the optimal control of noise and vibration is that an overlap point exists on the motion patterns of the flap with different frequencies.
- 8 February 2022
- Published by IOP Publishing in Smart Materials and Structures
- Vol. 31 (3) , 035023
Numerical and experimental analysis on the helicopter rotor dynamic load controlled by the actively trailing edge flap
Reducing the rotor dynamic load is an important issue to improve the performance and reliability of a helicopter. The control mechanism of the actively controlled flap (ACF) on the rotor dynamic load is numerically and experimentally investigated by a 3-blade helicopter rotor in this paper. In the aero-elastic numerical approach, the complex motion of the rotor such as the stretching, bending, torsion and pitching of the blade including the deflection of the ACF are all taken into consideration in the structural formulation. The aerodynamic solution adopted the vortex lattice method combined with the free wake model, in which the influence of ACF on the free wake and the aerodynamic load on the blade is taken into account as well. While the experimental method of measuring hub loads and acoustic was accomplished by a rotor rig in a wind tunnel. The result shows that the 3/rev ACF actuation can reduce the 3 ω hub load by more than 50% at maximum, which is significantly better than the 4/rev control. While 4/rev has greater potential to reduce blade vortex interaction (BVI) loads than 3/rev with µ = 0.15. Further mechanistic analysis shows that by changing the phase difference between the dynamic load on the flap and the rest of the blade, the peak load on the whole blade can be improved, thus achieving effective control of the hub dynamic load, the flap reaches the minimum angle of attack at 90 ∘ –100 ∘ azimuth under best control condition; when the BVI load is perfectly controlled, the flap reaches the minimum angle of attack at 140 ∘ azimuth, and by changing the circulation of the wake, the intensity of BVI in the advancing side is improved. Moreover, an interesting finding in the optimal control of noise and vibration is that an overlap point exists on the motion patterns of the flap with different frequencies.