Section: New Results
Trustworthy Collaboration
Participants : Claudia-Lavinia Ignat, Victorien Elvinger, François Charoy, Olivier Perrin, Gérald Oster, Hoang Long Nguyen.
Trust between users is an important factor for the success of a collaboration. Users might want to collaborate only with those users they trust. We are interested in assessing users trust according to their behaviour during collaboration in a large scale environment. We studied the trust assessment problem and designed a computational trust model for collaborative systems [1]. We also studied how to predict the trust relation between users that did not interact in the past. Given a network in which the links represent the trust/distrust relations between users, we aimed to predict future relations. We proposed a link-sign prediction algorithm [6] that does not require full graph information, is suitable for dynamic networks and takes into account the creation time of the links in the network. Our solution combines state-of-the-art techniques in natural language processing (Doc2Vec [25]) and deep learning (Recurrent Neural Networks [31] with Long-Short Term Memory [24]) with the random walk graph sampling [26]. Our algorithm outperforms state-of-the-art approaches on real world signed directed social network datasets. In distributed collaborative systems, participants maintain a replicated copy of shared documents. They edit their own copy and then share their modifications without any coordination. Copies follow successions of divergence and convergence. Convergence is a liveness property of collaborative systems. Some malicious participants may find an advantage to make the collaboration fail. To that end, they can preclude convergence of the copies. To protect convergence of copies, participants can exploit an authenticated log of modifications. New participants have to retrieve the entire log in order to contribute. Unfortunately, the cost of joining a collaboration increases with the size of this log. Causal Stability allows to prune authenticated logs in a static collaborative group without any malicious participants. We tailored Causal Stability to dynamic groups in the presence of malicious participants. We also proposed a mechanism to verify the consistency of a pruned log and a mechanism to authenticate a snapshot from a pruned log [7]. Public key server is a simple yet effective way of key management in secure end-to-end communication. To ensure the trustworthiness of a public key server, CONIKS [27] employs a tamper-evident data structure on the server and a gossiping protocol among clients in order to detect compromised servers. However, due to lack of incentive and vulnerability to malicious clients, a gossiping protocol is hard to implement in practice. Meanwhile, alternative solutions such as EthIKS [21] are too costly. We proposed Trusternity [13], [12], an auditing scheme relying on Ethereum blockchain that is easy to implement, inexpensive to operate and resilient to malicious clients. We also conducted an empirical study of system behaviour in face of attacks and proposed a lightweight anomaly detection algorithm to protect clients against such attacks.