Goren Gordon

Jonatan Barkan, Goren Gordon

Inspired by infants' intrinsic motivation to learn, which values informative sensory channels contingent on their immediate social environment, we developed a deep curiosity loop (DCL) architecture. The DCL is composed of a learner, which attempts to learn a forward model of the agent's state-action transition, and a novel reinforcement-learning (RL) component, namely, an Action-Convolution Deep Q-Network, which uses the learner's prediction error as reward. The environment for our agent is composed of visual social scenes, composed of sitcom video streams, thereby both the learner and the RL are constructed as deep convolutional neural networks. The agent's learner learns to predict the zero-th order of the dynamics of visual scenes, resulting in intrinsic rewards proportional to changes within its social environment. The sources of these socially informative changes within the sitcom are predominantly motions of faces and hands, leading to the unsupervised curiosity-based learning of social interaction features. The face and hand detection is represented by the value function and the social interaction optical-flow is represented by the policy. Our results suggest that face and hand detection are emergent properties of curiosity-based learning embedded in social environments.

Shahar Somin, Goren Gordon, Yaniv Altshuler

Blockchain technology, which has been known by mostly small technological circles up until recently, is bursting throughout the globe, with a potential economic and social impact that could fundamentally alter traditional financial and social structures. Issuing cryptocurrencies on top of the Blockchain system by startups and private sector companies is becoming a ubiquitous phenomenon, inducing the trading of these crypto-coins among their holders using dedicated exchanges. Apart from being a trading ledger for tokens, Blockchain can also be observed as a social network. Analyzing and modeling the dynamics of the "social signals" of this network can contribute to our understanding of this ecosystem and the forces acting within in. This work is the first analysis of the network properties of the ERC20 protocol compliant crypto-coins' trading data. Considering all trading wallets as a network's nodes, and constructing its edges using buy--sell trades, we can analyze the network properties of the ERC20 network. Examining several periods of time, and several data aggregation variants, we demonstrate that the network displays strong power-law properties. These results coincide with current network theory expectations, however nonetheless, are the first scientific validation of it, for the ERC20 trading data. The data we examined is composed of over 30 million ERC20 tokens trades, performed by over 6.8 million unique wallets, lapsing over a two years period between February 2016 and February 2018.