We presentHelix, a blockchain-based consensus protocol forfairordering of transactions among nodes in a distributed network. Helix advances in rounds, in each an elected primary node proposes a potential block (a successive set of transactions). For being included in the blockchain, a block must pass validation by an elected committee of nodes. Nodes have two primary preferences. First, to be elected as committee members. Additionally, because each transaction is associated with one of the network nodes, nodes would like to prioritize their own transactions over those of others. Our definition of fairness incorporates three key elements. First, the process of electing nodes to committees is random and unpredictable. Second, a correlated sampling scheme is used to guarantee random selection and ordering of pending transactions in blocks. Third, transactions are encrypted to hide their associations with nodes and prevent censorship. Through the corresponding threshold decryption process we obtain an unpredictable and non-manipulable randomness beacon, which serves both the election process and the correlated sampling scheme. We define a quantitative measure of fairness in the protocol, prove theoretically that fairness manipulation in Helix is significantly limited, and present experiments evaluating fairness in practice.

We provide theoretical investigation of curriculum learning in the context of stochastic gradient descent when optimizing the convex linear regression loss. We prove that the rate of convergence of an ideal curriculum learning method is monotonically increasing with the difficulty of the examples. Moreover, among all equally difficult points, convergence is faster when using points which incur higher loss with respect to the current hypothesis. We then analyze curriculum learning in the context of training a CNN. We describe a method which infers the curriculum by way of transfer learning from another network, pre-trained on a different task. While this approach can only approximate the ideal curriculum, we observe empirically similar behavior to the one predicted by the theory, namely, a significant boost in convergence speed at the beginning of training. When the task is made more difficult, improvement in generalization performance is also observed. Finally, curriculum learning exhibits robustness against unfavorable conditions such as excessive regularization.

We present Helix, a blockchain-based consensus protocol for fair ordering of transactions among nodes in a distributed network. Helix advances in rounds, where in each round, the primary node (elected among the network nodes) proposes a potential block (a successive set of transactions). In order to be included in the blockchain, a block must pass validation by an elected committee of nodes. Helix nodes are presumed to have two primary preferences. They prefer to be elected as committee members. Additionally, because each transaction is associated with one of the network nodes, they prefer to prioritize their own transactions over those of others. In light of these individual preferences, our definition of fairness incorporates three key elements. First, the process of electing nodes to committees is random and unpredictable. Second, a correlated sampling scheme is used in order to guarantee random selection and ordering of pending transactions in blocks. Third, transactions are encrypted in order to hide their associations with their respective nodes and prevent censorship. Through the corresponding threshold decryption process we obtain an unpredictable and non-manipulable randomness beacon, which serves both the election process and the correlated sampling scheme. We define a quantitative measure of fairness in the protocol, prove theoretically that fairness manipulation in Helix is significantly limited, and present experiments evaluating fairness in practice.

Studies in visual perceptual learning investigate the way human performance improves with practice, in the context of relatively simple (and therefore more manageable) visual tasks. Building on the powerful tools currently available for the training of Convolution Neural Networks (CNN), networks whose original architecture was inspired by the visual system, we revisited some of the open computational questions in perceptual learning. We first replicated two representative sets of perceptual learning experiments by training a shallow CNN to perform the relevant tasks. These networks qualitatively showed most of the characteristic behavior observed in perceptual learning, including the hallmark phenomena of specificity and its various manifestations in the forms of transfer or partial transfer, and learning enabling. We next analyzed the dynamics of weight modifications in the networks, identifying patterns which appeared to be instrumental for the transfer (or generalization) of learned skills from one task to another in the simulated networks. These patterns may identify ways by which the domain of search in the parameter space during network re-training can be significantly reduced, thereby accomplishing knowledge transfer.