# 2022FinTechathon 深圳国际金融科技大赛
### Fed-heathcare Team Won the 2022 FinTechathon Competition First Prize!
团队:Fed-healthcare (联邦医疗)
面向智能体音感知的联邦学习系统-基于FATE平台
![](/figures/2022FinTechathon1.jpg)
# FederatedHealth_HeartSound
The horizontal FL (HFL) and vertical FL (VFL) paradigms for heart sound analysis.
#### Index Terms— Computer audition, federated learning, heart sound, information security, model interpretability
# Algorithm_Models
* Paradigms and workflows of horizontal and vertical federated learning (FL) on multi-institutional heart sound databases.
#### **Note:** We first provide the final models and the relevant experimental results for this challenge. Other major program files will be uploaded after the paper of "Heart Sound Abnormality Detection from Multi-institutional Collaboration: Introducing a Federated Ensemble Learning Framework" is accepted.
## Results
* Horizontally-Federated Learning vs Data-Centralised Learning
Table 1. A SUMMARY OF RESULTS (IN [%]) FOR CLASSIC XGBOOST AND THE HFL MODEL WITH OPTIMAL PARAMETERS.
| | Acc | Se | Sp | UF1 | UAR |
| ----- | ----- | ---- |---- |---- |---- |
| XGBoost(Centralised Data)| 68.4 | 69.1 |67.6 | 68.4 | 68.4 |
| Homogeneous-SecureBoost | 67.5 | 62.1 |72.8 | 67.4 | 67.5 |
Important parameters settings for the HFL and the XGBoost: tree depth=3, tree number=30, subsample feature rate=1.0, learning rate=0.3.
![](/figures/HFL_matrix.jpg)
Fig. 1. Normalised confusion matrix (in [%]) for the XGBoost and Horizontal-SecureBoost models.
Table 2. COMPARISON OF THE RESULTS (IN [%]) OF THE CONVENTIONAL XGBOOST AND HETEROGENEOUS SECUREBOOST MODELS ON DATA FOR EACH INSTITUTION.
| | XGBoost(Centralised Data)| |Heterogeneous-SecureBoost| |
| | Acc | Se | Sp | UAR | Acc | Se | Sp | UAR |
| ----- | ----- | ---- |---- |---- |---- |----- | ---- |---- |
| Db | 86.7 |85.2 |88.3 |86.8 |82.7 |82.0 |83.5 |82.7 |
| Dc | 86.7 |85.7 |87.5 |86.6 |93.3 |85.7 |92.0 |92.9 |
| Dd | 93.3 |87.5 |92.0 |93.8 |96.2 |89.6 |96.4 |97.2 |
| De | 88.2 |85.6 |86.7 |87.8 |87.6 |82.6 |86.3 |84.3 |
| Df | 86.8 |90.9 |81.3 |86.1 |79.5 |75.5 |71.3 |78.4 |
![](/figures/VFL_matrix.jpg)
Fig. 2. Normalised confusion matrix (in [%]) for XGBoost and Vertically- SecureBoost models trained at D_e database.
## Awards
![](/figures/Awards11.png)
![](/figures/Awards22.png)
Fig. 3. 2022Fintechathon Shenzhen International FinTechathon Prize. https://www.infoq.cn/zones/fintechathon/campus2022/result
## Availability
1. Voice of the Body (VoB) 是第一个计算机听觉医学数据库平台,用于对体音信号进行分析. https://www.vob-bit.org/
2. Classification of Heart Sound Recordings (PhysioNet/CinC challenge): https://physionet.org/content/challenge-2016/1.0.0.
3. SHAP (SHapley Additive exPlanations) is a game-theoretic method to explain the output of ML models. https://shap.readthedocs.io.
4. FATE (Federated AI Technology Enabler) supports the FL architecture, as well as the secure computation and development of various ML algorithms. https://github.com/FederatedAI/FATE.
## Cite As
Wanyong Qiu, Chen Quan, Lixian Zhu, Yongzi Yu, Zhihua Wang, Yu Ma, Mengkai Sun, Yi Chang, Kun Qian*, Bin Hu∗, Yoshiharu Yamamoto and Bjoern W. Schuller, “Heart Sound Abnormality Detection from Multi-institutional Collaboration: Introducing a Federated Ensemble Learning Framework”, IEEE IoTJ, pp. 1-11, Submitted, July 2023.
没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
【项目资源】: 包含前端、后端、移动开发、操作系统、人工智能、物联网、信息化管理、数据库、硬件开发、大数据、课程资源、音视频、网站开发等各种技术项目的源码。 包括STM32、ESP8266、PHP、QT、Linux、iOS、C++、Java、python、web、C#、EDA、proteus、RTOS等项目的源码。 【项目质量】: 所有源码都经过严格测试,可以直接运行。 功能在确认正常工作后才上传。 【适用人群】: 适用于希望学习不同技术领域的小白或进阶学习者。 可作为毕设项目、课程设计、大作业、工程实训或初期项目立项。 【附加价值】: 项目具有较高的学习借鉴价值,也可直接拿来修改复刻。 对于有一定基础或热衷于研究的人来说,可以在这些基础代码上进行修改和扩展,实现其他功能。 【沟通交流】: 有任何使用上的问题,欢迎随时与博主沟通,博主会及时解答。 鼓励下载和使用,并欢迎大家互相学习,共同进步。
资源推荐
资源详情
资源评论
收起资源包目录
2022FinTechathon深圳国际金融科技大赛(人工智能赛道 ).zip (100个子文件)
loss_detail.csv 879B
model_summary.csv 224B
model_summary.csv 224B
xgb-HFL-matrix.eps 949KB
HFL-matrix.eps 942KB
xgb-VFL-matrix.eps 942KB
VFL-matrix.eps 931KB
fig3a-HFL.eps 746KB
fig3b-HFL.eps 719KB
fig3c-HFL.eps 715KB
fig3d-HFL.eps 679KB
fed-sum[abnormal].eps 231KB
global_sum[abnormal].eps 230KB
MMD_matrix.eps 137KB
sample-fed_[169].eps 59KB
sample_[169].eps 57KB
fed-bar[abnormal].eps 30KB
global_bar[abnormal].eps 25KB
Graphical_Abstract.jpg 144KB
shap1.jpg 94KB
shap2.jpg 42KB
HFL_results.jpg 36KB
2022FinTechathon1.jpg 34KB
MMD.jpg 23KB
HFL_matrix.jpg 19KB
VFL_matrix.jpg 19KB
README.md 4KB
fig1-HFL.pdf 199KB
fig2-VFL.pdf 183KB
fig10-VFL-alg.pdf 144KB
fig9-VFL-alg.pdf 112KB
fig8-HFL-alg.pdf 89KB
Awards1.png 240KB
Awards2.png 240KB
Awards11.png 193KB
Awards22.png 193KB
label0_id22_decision_tree.png 80KB
label0_id12_decision_tree.png 80KB
label0_id28_decision_tree.png 79KB
label0_id7_decision_tree.png 78KB
label0_id27_decision_tree.png 78KB
label0_id14_decision_tree.png 78KB
label0_id25_decision_tree.png 77KB
label0_id0_decision_tree.png 77KB
label0_id3_decision_tree.png 77KB
label0_id10_decision_tree.png 76KB
label0_id21_decision_tree.png 76KB
label0_id20_decision_tree.png 76KB
label0_id1_decision_tree.png 75KB
label0_id23_decision_tree.png 74KB
label0_id24_decision_tree.png 74KB
label0_id2_decision_tree.png 74KB
label0_id9_decision_tree.png 74KB
label0_id4_decision_tree.png 74KB
label0_id8_decision_tree.png 74KB
label0_id11_decision_tree.png 74KB
label0_id15_decision_tree.png 73KB
label0_id17_decision_tree.png 72KB
label0_id5_decision_tree.png 72KB
label0_id16_decision_tree.png 71KB
label0_id6_decision_tree.png 70KB
label0_id26_decision_tree.png 70KB
label0_id19_decision_tree.png 70KB
label0_id18_decision_tree.png 68KB
label0_id22_decision_tree.png 62KB
label0_id6_decision_tree.png 61KB
label0_id14_decision_tree.png 61KB
label0_id11_decision_tree.png 61KB
label0_id3_decision_tree.png 61KB
label0_id17_decision_tree.png 61KB
label0_id2_decision_tree.png 61KB
label0_id16_decision_tree.png 61KB
label0_id0_decision_tree.png 61KB
label0_id13_decision_tree.png 61KB
label0_id20_decision_tree.png 61KB
label0_id5_decision_tree.png 61KB
label0_id25_decision_tree.png 61KB
label0_id24_decision_tree.png 61KB
label0_id10_decision_tree.png 61KB
label0_id8_decision_tree.png 61KB
label0_id4_decision_tree.png 61KB
label0_id15_decision_tree.png 61KB
label0_id19_decision_tree.png 60KB
label0_id29_decision_tree.png 60KB
label0_id26_decision_tree.png 60KB
label0_id9_decision_tree.png 60KB
label0_id13_decision_tree.png 58KB
label0_id21_decision_tree.png 54KB
label0_id29_decision_tree.png 54KB
label0_id23_decision_tree.png 54KB
label0_id27_decision_tree.png 54KB
label0_id18_decision_tree.png 54KB
label0_id12_decision_tree.png 54KB
label0_id1_decision_tree.png 54KB
label0_id28_decision_tree.png 54KB
label0_id7_decision_tree.png 53KB
LOSS_train_loss.png 26KB
model_summary.png 6KB
model_summary.png 5KB
model_homo_secureboost.txt 22KB
共 100 条
- 1
资源评论
妄北y
- 粉丝: 1w+
- 资源: 1万+
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功