Daniel Cabarcas

Magali Bardet, Maxime Bros, Daniel Cabarcas et al.

Rank Decoding (RD) is the main underlying problem in rank-based cryptography. Based on this problem and quasi-cyclic versions of it, very efficient schemes have been proposed recently, such as those in the ROLLO and RQC submissions, which have reached the second round of the NIST Post-Quantum competition. Two main approaches have been studied to solve RD: combinatorial ones and algebraic ones. While the former has been studied extensively, a better understanding of the latter was recently obtained by Bardet et al. (EUROCRYPT20) where it appeared that algebraic attacks can often be more efficient than combinatorial ones for cryptographic parameters. This paper gives substantial improvements upon this attack in terms both of complexity and of the assumptions required by the cryptanalysis. We present attacks for ROLLO-I-128, 192, and 256 with bit complexity respectively in 70, 86, and 158, to be compared to 117, 144, and 197 for the aforementionned previous attack. Moreover, unlike this previous attack, ours does not need generic Gröbner basis algorithms since it only requires to solve a linear system. For a case called overdetermined, this modeling allows us to avoid Gröbner basis computations by going directly to solving a linear system. For the other case, called underdetermined, we also improve the results from the previous attack by combining the Ourivski-Johansson modeling together with a new modeling for a generic MinRank instance; the latter modeling allows us to refine the analysis of MinRank's complexity given in the paper by Verbel et al. (PQC19). Finally, since the proposed parameters of ROLLO and RQC are completely broken by our new attack, we give examples of new parameters for ROLLO and RQC that make them resistant to our attacks. These new parameters show that these systems remain attractive, with a loss of only about 50\% in terms of key size for ROLLO-I.

Niels Justesen, Miguel Gonzalez Duque, Daniel Cabarcas Jaramillo et al.

Imitation Learning (IL) is a machine learning approach to learn a policy from a dataset of demonstrations. IL can be useful to kick-start learning before applying reinforcement learning (RL) but it can also be useful on its own, e.g. to learn to imitate human players in video games. However, a major limitation of current IL approaches is that they learn only a single "average" policy based on a dataset that possibly contains demonstrations of numerous different types of behaviors. In this paper, we propose a new approach called Behavioral Repertoire Imitation Learning (BRIL) that instead learns a repertoire of behaviors from a set of demonstrations by augmenting the state-action pairs with behavioral descriptions. The outcome of this approach is a single neural network policy conditioned on a behavior description that can be precisely modulated. We apply this approach to train a policy on 7,777 human replays to perform build-order planning in StarCraft II. Principal Component Analysis (PCA) is applied to construct a low-dimensional behavioral space from the high-dimensional army unit composition of each demonstration. The results demonstrate that the learned policy can be effectively manipulated to express distinct behaviors. Additionally, by applying the UCB1 algorithm, we are able to adapt the behavior of the policy - in-between games - to reach a performance beyond that of the traditional IL baseline approach.