Federated Learning (FL)[1] is a type of distributed machine learning that allows the owners of the training data to preserve their privacy while still be- ing able to collectively train a model. FL is a new area in research and several chal- lenges reagarding privacy and communication cost still need to be overcome. Gradient leakage[1], for example is the possibility of partially reconstruct- ing the private data of a participant based on the weight gradients they send over the network dur- ing FL, which poses a great risk for privacy. Miti- gations against this problem lead to an increase in the computational complexity of the scheme or af- fect the performance of the resulting fully trained model. Other issues regarding trusting the central server or the clients, or accounting for clients that loose connection or drop out during training also exist. This paper is a literature survey about frameworks for Horizontal Federated Learning (HFL), which is a subset of Federated Learning, in which all clients have the same type of data (same set of features). The survey presents a summary of how 7 different Horizontal Federated Learning frameworks work, and compares them in terms of their performance and the security guarantees they provide. More- over, a summary of how each of the studied frame- works resolves the trade-offs among data privacy, framework performance and resulting model per- formance is also provided. Based on the studied frameworks it is concluded that the privacy and performance issues of HFL still need to be researched. Suggestions for future re- search topics are also provided.

Federated Learning starts to give a new perspective regarding the applicability of machine learning in real-life scenarios. Its main goal is to train the model while keeping the participants' data in their devices, thus guaranteeing the privacy of their data. One of the main architectures is the Horizontal Federated Learning, which is the most common one implemented. However, the challenges of the model (security attacks, data leakage), led to using some privacy-enhancing elements. Even those have their own trade-offs that make Federated Learning a challenge to implement in real-life scenarios (communication cost, training time). One question may rise up based on the aforementioned challenges: What if there is a hybrid way of implementing the Federated Learning so that we can overcome the challenges of the present implementations and expand its potential? This paper aims to answer this by diving into five hybrid models that use a variety of components for preserving privacy (Differential Privacy combined with Secure Multiparty Computation, blockchain) and will be compared with each other. Based on that, a reader can get an overview of how the hybrid approach affects the evolution of Horizontal Federated Learning.

Federated learning is a machine learning technique proposed by Google AI in 2016, as a solution to the GDPR regulations that made the classical Centralized Training, not only unfeasible, but also illegal, in some cases. In spite of its potential, FL has not gained much trust in the community, especially because of its susceptibility to data-privacy attacks. Incipient solutions to this problem were homomorphic encryption and differential privacy. These techniques, however, do not come with a solution to other open problems in the field, such as the difficulty to perform FL on vertically partitioned data, ensuring aggregation verifiability, resilience to user dropout etc. Fortunately, new strategies have been developed in the last years. This paper provides a comprehensive study of the state of the art privacy preserving techniques aimed to address two of these problems: vertical federated learning environments and aggregation verifiability. To this end, we study FedV, SecureBoost, MP-FEDXGB, VerifyNet, VFL, together with a verifiability approach exploiting bilinear aggregate signatures, analysing their security model, complexity and communication overhead, the accuracy impact, benefits and downsides in a comparative manner.

Privacy in today's world is a very important topic and all the more important when sizeable amounts of data are needed in Neural Network processing models. Federated Learning is a technique which aims to decentralize the training process in order to allow the clients to maintain their privacy, while also contributing to a broader learning process. In order to allow parties that undertake similar tasks to share data between them, even if they don't follow the same feature representation or domain distribution, Transfer Learning is also used in order to augment the learning by sharing knowledge with the contributing parties. The name of this combination of techniques is Federated Transfer Learning. This paper aims to showcase the strengths and weaknesses of Federated Learning through a simple implementation while comparing different Federated Transfer Learning frameworks that can be used in order to enhance the capabilities of a simple federation of clients that are contributing towards the learning of a similar task.