Arrow Research search

Author name cluster

Chenyang Jiang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers
1 author row

Possible papers

2

YNIMG Journal 2025 Journal Article

A large-scale multi-centre study characterising atrophy heterogeneity in Alzheimer’s disease

  • Vikram Venkatraghavan
  • Damiano Archetti
  • Pierrick Bourgeat
  • Chenyang Jiang
  • Mara ten Kate
  • Anna C. van Loenhoud
  • Rik Ossenkoppele
  • Charlotte E. Teunissen

Previous studies identified atrophy-based Alzheimer's disease(AD) subtypes linked to distinct clinical symptoms, but their consistency across subtyping approaches remains unclear. This large-scale study evaluates subtype concordance using two data-driven approaches. In this work, we analyzed data from n=10,011 patients across 10 AD cohorts spanning Europe, the US, and Australia, extracting regional volumes using Freesurfer. To characterize atrophy heterogeneity in the AD continuum, we developed a two-step approach, Snowphlake (Staging NeurOdegeneration With PHenotype informed progression timeLine of biomarKErs), to identify subtypes and atrophy-event sequences within each subtype. Results were compared with SuStaIn (Subtype and Stage Inference), which jointly estimates subtypes and staging, using similar training and validation. Training included Aβ+ participants (n=1,195) and Aβ- cognitively unimpaired controls (n=1,692). We validated model-staging in a held-out clinical dataset (n=6,362) and an independent dataset (n=762), and assessed clinical significance in Aβ+ subsets(n=1,796 held-out; n=159 external). Concordance analysis evaluated consistency between methods. In the AD dementia(AD-D) training data, both Snowphlake and SuStaIn identified four subtypes. In the validation datasets, staging with both methods correlated with Mini-Mental State Examination(MMSE) scores. The Snowphlake subtypes assigned in Aβ+ validation datasets were associated with alterations in specific cognitive domains(Cohen's f:[0.15-0.33]). Similarly, the SuStaIn subtypes were also associated specific cognitive domains(Cohen's f:[0.17-0.34]). However, we observed low concordance between Snowphlake and SuStaIn, with 39.7% of AD-D patients grouped in concordant subtypes by both methods. In conclusion, Snowphlake and SuStaIn identified four atrophy-based subtypes that linked to distinct symptom profiles. While this highlights that the neuro-anatomically defined subtypes also meaningfully associate with different cognitive impairments at a group level, the low concordance between methods suggests that future research is needed to better understand the biological and methodological factors contributing to the observed variability.

Details DOI

NeurIPS Conference 2025 Conference Paper

Rectifying Soft-Label Entangled Bias in Long-Tailed Dataset Distillation

  • Chenyang Jiang
  • Hang Zhao
  • Xinyu Zhang
  • Zhengcen Li
  • Qiben Shan
  • Shaocong Wu
  • Jingyong Su

Dataset distillation compresses large-scale datasets into compact, highly informative synthetic data, significantly reducing storage and training costs. However, existing research primarily focuses on balanced datasets and struggles to perform under real-world long-tailed distributions. In this work, we emphasize the critical role of soft labels in long-tailed dataset distillation and uncover the underlying mechanisms contributing to performance degradation. Specifically, we derive an imbalance-aware generalization bound for model trained on distilled dataset. We then identify two primary sources of soft-label bias, which originate from the distillation model and the distilled images, through systematic perturbation of the data imbalance levels. To address this, we propose ADSA, an Adaptive Soft-label Alignment module that calibrates the entangled biases. This lightweight module integrates seamlessly into existing distillation pipelines and consistently improves performance. On ImageNet-1k-LT with EDC and IPC=50, ADSA improves tail-class accuracy by up to 11. 8\% and raises overall accuracy to 41. 4\%. Extensive experiments demonstrate that ADSA provides a robust and generalizable solution under limited label budgets and across a range of distillation techniques.

PDF Details