The use of causal effect estimation methods to estimate treatment outcomes from observational studies is widespread in the field of medicine, social sciences and econometrics. One of the main challenges in this field is data scarcity. Although these methods rely heavily on highly sensitive information, so far there has been a huge lack of privacy preserving approaches for the same. Another challenge is about generalization capacity of the causal inference model trained on different source domains for a specific case since the counterfactuals are never observed as well as the possibility of non identically distributed data in source/target domains. In this paper, we aim to tackle these challenges with the help of a federated invariant learning framework. Our approach can help exploit additional data sources to facilitate privacy-preserving causal effect estimation in an unseen target population. We evaluate our proposed framework on synthetic and semi-synthetic datasets and show that the empirical results with our distributed approaches are almost consistent with the current centralized approaches to treatment effect estimation with added advantage of better generalization.

Federated learning has become increasingly popular as it facilitates collaborative training among multiple clients under the coordination of a central server, while preserving the client data privacy. In practice, some of the main challenges posed to federated learning is model robustness, out of distribution generalization, fair incentivization and vulnerability to various privacy attacks. We propose a novel blockchain based causal approach to developing a robust federated ecosystem that achieves generalization beyond the limited set of client environments accessible to the server during a specific round of training, with the capability to extrapolate to new and unseen environments in server/client side enhancing fair incentivization for all participants. For real world deployment of federated systems, participant incentivization based on causal/invariant learning as opposed to associational learning methods will prove to be extremely beneficial in terms of fairness, privacy and robustness.

Federated Learning is an emerging privacy-preserving distributed machine learning approach to building a shared model by performing distributed training locally on participating devices (clients) and aggregating the local models into a global one. As this approach prevents data collection and aggregation, it helps in reducing associated privacy risks to a great extent. However, the data samples across all participating clients are usually not independent and identically distributed (non-iid), and Out of Distribution(OOD) generalization for the learned models can be poor. Besides this challenge, federated learning also remains vulnerable to various attacks on security wherein a few malicious participating entities work towards inserting backdoors, degrading the generated aggregated model as well as inferring the data owned by participating entities. In this paper, we propose an approach for learning invariant (causal) features common to all participating clients in a federated learning setup and analyze empirically how it enhances the Out of Distribution (OOD) accuracy as well as the privacy of the final learned model.

Federated Averaging (FedAVG) has become the most popular federated learning algorithm due to its simplicity and low communication overhead. We use simple examples to show that FedAVG has the tendency to sew together the optima across the participating clients. These sewed optima exhibit poor generalization when used on a new client with new data distribution. Inspired by the invariance principles in (Arjovsky et al., 2019; Parascandolo et al., 2020), we focus on learning a model that is locally optimal across the different clients simultaneously. We propose a modification to FedAVG algorithm to include masked gradients (AND-mask from (Parascandolo et al., 2020)) across the clients and uses them to carry out an additional server model update. We show that this algorithm achieves better accuracy (out-of-distribution) than FedAVG, especially when the data is non-identically distributed across clients.

We want to take advantage of the different environments (growth conditions, cell lines) present in gene expression datasets to get a better insight into the actual mechanisms that happen inside the cell. We would like to avoid relying on any explicit prior knowledge such as pathways that are always incomplete.We propose a multi-environment training procedure that aims at learning cell embeddings which are disentangled from the drug effect point of view. Our model is similar to a Conditional VAE along with an attention mechanism that can sparsely modify the prior distribution in latent space based on the environment.

No current algorithm or network can learn all the possible shapes in a scene. In this project, we take on the task of explicit representation by predicting the configuration of a graph of features with an energy network. Using both the information encoded in the vertices and edges of the graph of simple geometric features, we find what shapes arise in a point cloud. We've decided to work on point cloud as it the most difficult setting, and most methods in it can be generalized to one with more information. Unlike images, semantic learning on 3D point clouds using a deep network is challenging due to the natural way data is unstructured. Hence we aim to do graph partitioning with the goal of finding the lowest cost unions, but where the result of alterations is unknown unless we compute the energy. But also with the possibility of edge additions. In a sense like a flow of mixed elements: if we let a node flow into a set, that element may "react" to increase or dampen the energy. Finally, we want unsupervised learning as most use cases have no ground truth; we simply want the best solution. We hope the network architecture presented here will the reader's interest as much it did ours.

Although Federated Learning allows for participants to contribute their local data without it being revealed, it faces issues in data security and in accurately paying participants for quality data contributions. In this paper, we propose an EOS Blockchain design and workflow to establish data security, a novel validation error based metric upon which we qualify gradient uploads for payment, and implement a small example of our blockchain Federated Learning model to analyze its performance.

Everyone wants to make better decisions. The impact of a decision on an outcome of interest is called a causal effect, and is traditionally estimated by performing randomized experiments. However, large data sources such as electronic medical records present opportunities to study causal effects of interventions that are difficult to evaluate through experiments. One example is the management of septic patients in the ICU. This typically involves performing several interventions in sequence, the choice of one depending on the outcome of others. Successfully evaluating the effect of these choices depends on strong assumptions, such as having adjusted for all confounding variables. While many argue that having high-dimensional data increases the likelihood of this assumption being true, it also introduces new challenges: the more variables we use for estimating effects, the less likely that patients who received different treatments are similar in all of them. In this talk, we will discuss causal effect estimation and treatment group overlap. We will also discuss the potential outcomes framework, classical methods for estimating causal effects, as well as new ones, tailored for working with large datasets.

Analysis of ways to reduce execution speed of Tiny YOLO V1 in flightkit GPU with an in depth PrecisionLoss comparison of Tiny YOLO V1 32 to that of 16Bit.

YOLO execution speed improvement - optimization steps as well as issues faced