Yuexin Ma
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
Research Subjects
Yuexin Ma has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Chengpan Tang
Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, China
2 shared publications - Jianhua Yang
Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, China
2 shared publications - Xiangwei Zhu
Sun Yat-sen University
2 shared publications - Xiaogong Hu
Shanghai Astronomical Observatory, Chinese Academy of Sciences, Shanghai 200030, China
2 shared publications - Yufei Yang
Beijing Satellite Navigation Center, Beijing 100094, China
2 shared publications Chunying Wang
State Key Laboratory of Wheat Improvement Shandong Agricultural University Taian Shandong China
1 shared publication
Latest Publications
- Open Access
- 27 February 2026
- Published by Wiley in Plant Biotechnology Journal
- Vol. 24 (6) , 4005-4026
DNAwhisper : An Integrated Deep Learning Pyramidal Framework for Multi‐Trait Genomic Prediction and Adaptive Marker Prioritisation
Genomic selection (GS) is critical for accelerating genetic gain in modern plant breeding. Deep learning approaches offer powerful non-linear representation capabilities for modelling non-additive effects. However, their application in GS remains restricted, as high-dimensional, low-sample and noisy data hinder the identification of informative markers. The present study proposes DNAwhisper, a deep learning framework designed for multi-trait prediction and adaptive marker prioritisation. The framework integrates a cascaded architecture, GFIformer, employing shared network parameters across partitioned marker blocks to adaptively compress genetic features within a hierarchical pyramid. Pre-training on population genetic structure regularises feature learning to establish a generalisable latent representation. During trait modelling, importance scores for aggregated genomic regions at multi-resolutions are extracted from the distinct pyramid levels under trait-guided deep supervision, enhancing interpretability and supporting marker prioritisation. DNAwhisper was evaluated on maize, wheat, tomato and grape datasets for marker prioritisation and phenotypic prediction, achieving prediction accuracy approximately 3.0% to 10.0% higher than the baseline model. Furthermore, DNAwhisper identifies major QTLs (e.g., , ) and epistatic signals within the gibberellin metabolic pathway across maize flowering traits. This framework provides a new strategy for dissecting the genetic architecture of complex traits.- 27 February 2026
- Published by Wiley in Plant Biotechnology Journal
- Vol. 24 (6) , 4005-4026
Open AccessDNAwhisper : An Integrated Deep Learning Pyramidal Framework for Multi‐Trait Genomic Prediction and Adaptive Marker Prioritisation
Genomic selection (GS) is critical for accelerating genetic gain in modern plant breeding. Deep learning approaches offer powerful non-linear representation capabilities for modelling non-additive effects. However, their application in GS remains restricted, as high-dimensional, low-sample and noisy data hinder the identification of informative markers. The present study proposes DNAwhisper, a deep learning framework designed for multi-trait prediction and adaptive marker prioritisation. The framework integrates a cascaded architecture, GFIformer, employing shared network parameters across partitioned marker blocks to adaptively compress genetic features within a hierarchical pyramid. Pre-training on population genetic structure regularises feature learning to establish a generalisable latent representation. During trait modelling, importance scores for aggregated genomic regions at multi-resolutions are extracted from the distinct pyramid levels under trait-guided deep supervision, enhancing interpretability and supporting marker prioritisation. DNAwhisper was evaluated on maize, wheat, tomato and grape datasets for marker prioritisation and phenotypic prediction, achieving prediction accuracy approximately 3.0% to 10.0% higher than the baseline model. Furthermore, DNAwhisper identifies major QTLs (e.g., , ) and epistatic signals within the gibberellin metabolic pathway across maize flowering traits. This framework provides a new strategy for dissecting the genetic architecture of complex traits. - 5 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (11) , 9510-9517
A New Representation of Universal Successor Features for Enhancing the Generalization of Target-Driven Visual Navigation
Target-driven visual navigation is a long-standing objective in the field of robotics. Deep reinforcement learning methods have demonstrated their effectiveness in developing target-driven visual navigation policies, yet they often struggle with generalization. Although extended reinforcement learning methods based on successor features enhance cross-target generalization, they falter in unseen environments. To tackle this issue, we put forth an approach that builds upon successor features and introduces two novel designs: the parsimonious dynamics and the adaptive policy. Specifically, we employ state features that predict successor features to create latent parsimonious dynamics models. Furthermore, learning a set of rules or dynamics within the latent state facilitates the identification of informative features for prediction and the acquisition of their corresponding successor features. The adaptive policy is to prevent the agent getting trapped by repeatedly taking the same action repeatedly even though it doesn't lead to any progress. Comprehensive results from the interactive AI2-THOR and the Habitat simulator show that agents trained using this method achieve state-of-the-art capabilities in generalization across target, scene, and domain. Additionally, we propose a continuous learning experiment that utilizes pre-trained models to achieve superior generalization.
- 5 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (11) , 9510-9517
A New Representation of Universal Successor Features for Enhancing the Generalization of Target-Driven Visual Navigation
Target-driven visual navigation is a long-standing objective in the field of robotics. Deep reinforcement learning methods have demonstrated their effectiveness in developing target-driven visual navigation policies, yet they often struggle with generalization. Although extended reinforcement learning methods based on successor features enhance cross-target generalization, they falter in unseen environments. To tackle this issue, we put forth an approach that builds upon successor features and introduces two novel designs: the parsimonious dynamics and the adaptive policy. Specifically, we employ state features that predict successor features to create latent parsimonious dynamics models. Furthermore, learning a set of rules or dynamics within the latent state facilitates the identification of informative features for prediction and the acquisition of their corresponding successor features. The adaptive policy is to prevent the agent getting trapped by repeatedly taking the same action repeatedly even though it doesn't lead to any progress. Comprehensive results from the interactive AI2-THOR and the Habitat simulator show that agents trained using this method achieve state-of-the-art capabilities in generalization across target, scene, and domain. Additionally, we propose a continuous learning experiment that utilizes pre-trained models to achieve superior generalization.
- 1 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-13
Discussions of a2-Drift Variations of BeiDou-3 Satellite Rubidium Atomic Clocks
Slow and significant stochastic variations named “a2-drift” variations are observed in this contribution, which are discovered during the estimation of frequency aging of BeiDou-3 rubidium (Rb) clocks. The magnitude of the variations suffered on BeiDou-3 Rbs is distinctly larger than GPS and appears to exhibit a systematic drift associated with time. The phenomenon is widely ignored and will compromise the long-term (one day and beyond) autonomy of the onboard clock units. Random run frequency modulation (RRFM) noise, which may induce similar variations, can be detected and identified by lag 1 autocorrelation function (lag 1 ACF) and Hadamard deviation. To retrace the origin of the variations, the Hadamard deviations, broadcast clock model parameter $a1$ , and frequency drift of BeiDou-3 Rbs are estimated with reference to GPS. The variations result in long-term fluctuations on the broadcast clock parameters ( $a1$ and $a2$ ) and arise an increase in the Hadamard deviation when the averaging time is greater than 1e 5 s. Furthermore, the simulations of the power-law noises reproduce the variations, and the lag 1 ACF supports that the variations are caused by RRFM. Besides this, many conclusions are also related to this variation. Using two years of multi-global navigation satellite system (GNSS) experiment (MGEX) clock offsets from 2019 to 2021, the following conclusions are drawn: 1) the RRFM emerges as dominant noise and causes a decline in the frequency stability when the averaging time is greater than 1e 5 s; 2) the variations on the estimated series of BeiDou-3 Rbs broadcast clock model $a1$ and frequency drift are more visible than those of GPS; 3) when the averaging time is approximately one day, the frequency stability of BeiDou-3 Rbs is similar to that of GPS; however, when the averaging time reaches 1e 6 s, the frequency stability of BeiDou-3 Rbs deteriorates because of the RRFM and is nearly ten times lower than that of GPS Rbs; and 4) the GPS Rbs are free from RRFM, but the BeiDou-3 Rbs contain significant features of random-walk frequency modulation (RWFM) and RRFM noise.
- 1 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-13
Discussions of a2-Drift Variations of BeiDou-3 Satellite Rubidium Atomic Clocks
Slow and significant stochastic variations named “a2-drift” variations are observed in this contribution, which are discovered during the estimation of frequency aging of BeiDou-3 rubidium (Rb) clocks. The magnitude of the variations suffered on BeiDou-3 Rbs is distinctly larger than GPS and appears to exhibit a systematic drift associated with time. The phenomenon is widely ignored and will compromise the long-term (one day and beyond) autonomy of the onboard clock units. Random run frequency modulation (RRFM) noise, which may induce similar variations, can be detected and identified by lag 1 autocorrelation function (lag 1 ACF) and Hadamard deviation. To retrace the origin of the variations, the Hadamard deviations, broadcast clock model parameter $a1$ , and frequency drift of BeiDou-3 Rbs are estimated with reference to GPS. The variations result in long-term fluctuations on the broadcast clock parameters ( $a1$ and $a2$ ) and arise an increase in the Hadamard deviation when the averaging time is greater than 1e 5 s. Furthermore, the simulations of the power-law noises reproduce the variations, and the lag 1 ACF supports that the variations are caused by RRFM. Besides this, many conclusions are also related to this variation. Using two years of multi-global navigation satellite system (GNSS) experiment (MGEX) clock offsets from 2019 to 2021, the following conclusions are drawn: 1) the RRFM emerges as dominant noise and causes a decline in the frequency stability when the averaging time is greater than 1e 5 s; 2) the variations on the estimated series of BeiDou-3 Rbs broadcast clock model $a1$ and frequency drift are more visible than those of GPS; 3) when the averaging time is approximately one day, the frequency stability of BeiDou-3 Rbs is similar to that of GPS; however, when the averaging time reaches 1e 6 s, the frequency stability of BeiDou-3 Rbs deteriorates because of the RRFM and is nearly ten times lower than that of GPS Rbs; and 4) the GPS Rbs are free from RRFM, but the BeiDou-3 Rbs contain significant features of random-walk frequency modulation (RWFM) and RRFM noise.
- 25 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-15
Long-Term Frequency Stability Analysis for RRFM and Power-Law Noise Determination of the BeiDou-3 Satellite On-Board Atomic Clocks
The key work this study is the power-law noises determination of 18 BeiDou-3 satellites (PRN19-PRN30 and PRN32-PRN37) on-board atomic clocks. First, for the noise determination of white frequency modulation (WFM), random-walk FM (RWFM), and random-run FM (RRFM), which are the basis of the ensemble timescale filter (ETF), we evaluated the Hadamard deviation with a maximum averaging time up to 30 days. To ensure the reliability of the evaluation, the Hadamard deviation was calculated by the relative clock offset determined from two independent techniques: orbit determination and time synchronization (ODTS) and the intersatellite link (ISL). This work shows that the noise characteristics of the clock offset obtained from ODTS and ISL are different, ODTS clocks are conducive to WFM determination because of low observation noise, and the ISL clocks are conducive to RWFM and RRFM determination because of independent to satellite orbit. Second, the lag 1 and $B1$ bias functions were employed to identify different noises, and the intensities of WFM, RWFM, and RRFM were determined by the Hadamard deviations above. The results of this study indicate that: 1) rubidium clocks (Rbs) always suffer from significant RRFM noise, which improves the magnitude component noise when the averaging time achieves several days; 2) the Rbs of BeiDou-3 satellites have distinctive characteristics of WFM, RWFM, and RRFM, different from the passive hydrogen masers (PHMs), which can only discover the WFM and RWFM noises; and 3) because of the RRFM, when the averaging time is greater than 2e5 s, the Hadamard deviation of Rbs is increased, which causes the frequency stability of Rbs to become an order of magnitude lower than PHMs. Last, simulation analyses and natural Kalman timescale (NKT) constructions using the actual ISL data were employed to verify the effectiveness of noises determination. Through this work, we can almost certainly identify the process noise of ETFs, which can become a reference for ETF construction in the future.
- 25 February 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-15
Long-Term Frequency Stability Analysis for RRFM and Power-Law Noise Determination of the BeiDou-3 Satellite On-Board Atomic Clocks
The key work this study is the power-law noises determination of 18 BeiDou-3 satellites (PRN19-PRN30 and PRN32-PRN37) on-board atomic clocks. First, for the noise determination of white frequency modulation (WFM), random-walk FM (RWFM), and random-run FM (RRFM), which are the basis of the ensemble timescale filter (ETF), we evaluated the Hadamard deviation with a maximum averaging time up to 30 days. To ensure the reliability of the evaluation, the Hadamard deviation was calculated by the relative clock offset determined from two independent techniques: orbit determination and time synchronization (ODTS) and the intersatellite link (ISL). This work shows that the noise characteristics of the clock offset obtained from ODTS and ISL are different, ODTS clocks are conducive to WFM determination because of low observation noise, and the ISL clocks are conducive to RWFM and RRFM determination because of independent to satellite orbit. Second, the lag 1 and $B1$ bias functions were employed to identify different noises, and the intensities of WFM, RWFM, and RRFM were determined by the Hadamard deviations above. The results of this study indicate that: 1) rubidium clocks (Rbs) always suffer from significant RRFM noise, which improves the magnitude component noise when the averaging time achieves several days; 2) the Rbs of BeiDou-3 satellites have distinctive characteristics of WFM, RWFM, and RRFM, different from the passive hydrogen masers (PHMs), which can only discover the WFM and RWFM noises; and 3) because of the RRFM, when the averaging time is greater than 2e5 s, the Hadamard deviation of Rbs is increased, which causes the frequency stability of Rbs to become an order of magnitude lower than PHMs. Last, simulation analyses and natural Kalman timescale (NKT) constructions using the actual ISL data were employed to verify the effectiveness of noises determination. Through this work, we can almost certainly identify the process noise of ETFs, which can become a reference for ETF construction in the future.