- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper investigates the knowledge-sharing problem in medical federated multi-task learning (FMTL), where transferable and task-specific knowledge are coupled within local task models, potentially causing negative transfer during aggregation.
The authors propose IPL-FL, a framework that decouples cross-client knowledge sharing from local task models through input-space perturbation generators, incorporating task-guided perturbation learning (TGPL) and relation-aware personalized aggregation (RAPA).
They evaluate the approach across eight clients spanning medical segmentation, classification, and detection tasks with heterogeneous model architectures, reporting a 5.65% average relative improvement over local training and outperforming existing FMTL baselines.
The study contributes an input-space knowledge-sharing framework for heterogeneous FMTL and an aggregation mechanism based on update-direction consistency and conditional information gain.
Critical strengths
- The paper identifies the knowledge-selection conflict in heterogeneous FMTL and provides empirical evidence of negative transfer through cross-task encoder aggregation experiments.
- IPL-FL separates task-specific modeling from federated knowledge sharing through TGPL and RAPA, enabling collaboration across heterogeneous tasks and model architectures without directly aggregating local predictors.
- The experiments cover eight clients across three medical task categories, with comparisons against heterogeneous FMTL methods and aggregation baselines demonstrating consistent improvements over local training and additional gains from RAPA.
Critical weaknesses
- The low-rank decomposition in RAPA assumes that shared subspaces capture transferable knowledge while residuals preserve client-specific information, but the experiments do not directly validate this distinction.
- Although the paper evaluates different perturbation configurations, it does not isolate the contributions of RAPA's low-rank decomposition, directional consistency, conditional information gain, and residual preservation
Questions for authors
see weaknesses
2 comments
Thank you for the detailed rebuttal and additional experiments. I think the new ablation studies have addressed my concerns about the knowledge decoupling assumption and the contributions of different components in RAPA. Therefore, I will keep my original score unchanged.
Our Response to Reviewer dT1T
We thank the reviewer for your constructive comment.
A1: To empirically validate our knowledge decoupling assumption, we conducted an ablation study evaluating what happens when different decomposed components are aggregated across clients—specifically comparing aggregating only the residual component (E), only the low-rank component (Upubr), and both components (Upubr+E):
a. Ablation of the low-rank component Upubr and the residual component E (Table 1)
As shown in Table 1, aggregating only the residual component (E) yields a Δm of +1.95%, which is 0.57% lower than the non-aggregated Local+TGPL baseline (+2.52%). This performance degradation confirms that E primarily captures task-specific information whose cross-client sharing causes negative transfer. In contrast, aggregating only the low-rank component (Upubr) achieves the best performance (+5.65%), demonstrating that Upubr isolates generalizable, transferable knowledge. Furthermore, although aggregating both components (Upubr+E) improves Δm by 0.83% over Local+TGPL (+3.35% vs. +2.52%), it still underperforms aggregating Upubr alone by 2.30% due to interference from the task-specific residual. These findings directly validate our decoupling design.
Manuscript Revision: We have incorporated this component ablation study and analysis into Page 9 of the Main Paper.
A2: To isolate the individual contributions of all four components in RAPA, we conducted a comprehensive ablation study.
a. Ablation of the directional consistency (aij) and conditional information gain (gj∣Si) (Table 2)
The results in Table 2 show that when both components are removed (aij=1,gj∣Si=1), performance drops to +2.76%, which is nearly identical to non-aggregated Local+TGPL (+2.52%). Between the two mechanisms, directional consistency exhibits a stronger individual impact: retaining only directional consistency (aij=1,gj∣Si=1) achieves a Δm of +4.25%, compared to +3.77% when retaining only conditional information gain (aij=1,gj∣Si=1). This aligns with our intuition that filtering out conflicting knowledge is more critical than pruning redundant information in heterogeneous multi-task settings, while combining both mechanisms yields the full +5.65% improvement.
Manuscript Revision: We have incorporated directional consistency (aij) and conditional information gain (gj∣Si) ablation results into Page 9 of the Main Paper.