Federated Learning

Federated learning allows models to complete training updates locally on the data, uploading only gradients rather than raw data

Federated learning allows multiple agencies to jointly train a model without sharing raw data, with each agency's data remaining local and only model gradients being exchanged.

Federated learning was proposed by Google in 2016, keeping data local and uploading only model gradients

Privacy-preserving technologies such as federated learning and on-device inference allow models to process data on the user's device, avoiding the upload of raw personal data to the cloud.

Federated learning was proposed by Google in 2016, originally to address privacy issues in model training on mobile devices.

Federated learning faces challenges including low communication efficiency, convergence difficulties caused by data heterogeneity, and gradient leakage attacks.

Differential privacy makes it impossible for attackers to reverse-engineer any individual user's data from the output by adding mathematically calibrated random noise to training data or model outputs.

Google has applied Federated Learning technology to Gboard keyboard prediction.

Privacy-enhancing technologies (PETs) include homomorphic encryption, federated learning, differential privacy and secure multi-party computation, with the goal of maintaining privacy protection at the data usage stage as well, but they still face deployment challenges such as high performance overhead and high engineering complexity.

In real clinical scenarios, federated learning and differential privacy face implementation obstacles such as differences in data distribution across institutions (non-IID), communication overhead and the complexity of compliance auditing, and there is still a distance to large-scale clinical deployment.

The research presents four typical trade-offs between privacy and performance: latency vs privacy, model accuracy vs differential privacy, scalability vs federated learning, and transparency vs proprietary SDKs.

DP-SGD is a mechanism for achieving differential privacy in stochastic gradient descent, implemented by clipping per-sample gradients and then adding Gaussian noise.

Federated learning lets multiple devices train collaboratively without centralizing data; volunteer computing projects such as BOINC and Folding@home have long aggregated idle CPU/GPU power for scientific computing.

Communication rounds are a key metric for evaluating the practicality of federated learning systems; the fewer rounds needed to reach a target accuracy, the higher the system's usability and cost-effectiveness, especially in resource-constrained scenarios such as edge devices and mobile endpoints.

Federated learning can train a shared global model from data distributed across clients without centralizing the raw data.

When statistical heterogeneity exists in the data distributions of clients, the global model tends to cater to clients with larger data volumes and mainstream distributions, causing clients that deviate from the mainstream to exhibit significantly worse local accuracy.

Existing fairness-oriented multi-objective federated optimization methods address the problem primarily through three approaches: client reweighting, gradient correction, and common descent.

The framework uses a federated learning mechanism, allowing multiple vehicles or ECUs to collaboratively train a shared model without exchanging raw CAN data.

The basic process of federated learning involves a central server distributing a global model, local nodes training on local data and uploading only gradients or model weight differences, and the server aggregating the updates via algorithms such as FedAvg before distributing a new global model.

Federated learning faces challenges: model updates may leak statistical features of training data through gradient inversion attacks, and large distribution differences across vehicle data can cause non-IID problems that lead to convergence difficulties or performance degradation.