We prove that for every constant c and epsilon = (log n)^{-c}, there is no polynomial time algorithm that when given an instance of 3-LIN with n variables where an (1 - epsilon)-fraction of the clauses are satisfiable, finds an assignment that satisfies atleast (1/2 + epsilon)-fraction of clauses unless NP subseteq BPP. The previous best hardness using a polynomial time reduction achieves epsilon = (log log n)^{-c}, which is obtained by the Label Cover hardness of Moshkovitz and Raz [J. ACM, 57(5), 2010] followed by the reduction from Label Cover to 3-LIN of Håstad [J. ACM, 48(4):798 - 859, 2001]. Our main idea is to prove a hardness result for Label Cover similar to Moshkovitz and Raz where each projection has a linear structure. This linear structure of Label Cover allows us to use Hadamard codes instead of long codes, making the reduction more efficient. For the hardness of Linear Label Cover, we follow the work of Dinur and Harsha [SIAM J. Comput., 42(6):2452 - 2486, 2013] that simplified the construction of Moshkovitz and Raz, and observe that running their reduction from a hardness of the problem LIN (of unbounded arity) instead of the more standard problem of solving quadratic equations ensures the linearity of the resultant Label Cover.

We study the problem of efficient online multiclass linear classification with bandit feedback, where all examples belong to one of $K$ classes and lie in the $d$-dimensional Euclidean space. Previous works have left open the challenge of designing efficient algorithms with finite mistake bounds when the data is linearly separable by a margin $\gamma$. In this work, we take a first step towards this problem. We consider two notions of linear separability: strong and weak. 1. Under the strong linear separability condition, we design an efficient algorithm that achieves a near-optimal mistake bound of $O\left(\frac{K}{\gamma^2} \right)$. 2. Under the more challenging weak linear separability condition, we design an efficient algorithm with a mistake bound of $2^{\widetilde{O}(\min(K \log^2 \frac{1}{\gamma}, \sqrt{\frac{1}{\gamma}} \log K))}$. Our algorithm is based on kernel Perceptron, which is inspired by the work of Klivans & Servedio (2008) on improperly learning intersection of halfspaces.

In the simultaneous Max-Cut problem, we are given k weighted graphs on the same set of n vertices, and the goal is to find a cut of the vertex set so that the minimum, over the k graphs, of the cut value is as large as possible. Previous work [BKS15] gave a polynomial time algorithm which achieved an approximation factor of 1/2 – o(1) for this problem (and an approximation factor of 1/2 + εk in the unweighted case, where εk → 0 as k → ∞). In this work, we give a polynomial time approximation algorithm for simultaneous Max-Cut with an approximation factor of 0.8780 (for all constant k). The natural SDP formulation for simultaneous Max-Cut was shown to have an integrality gap of 1/2 + εk in [BKS15]. In achieving the better approximation guarantee, we use a stronger Sum-of-Squares hierarchy SDP relaxation and a rounding algorithm based on Raghavendra-Tan [RT12], in addition to techniques from [BKS15].

A Boolean function f:{0,1}^n\->{0,1} is called a dictator if it depends on exactly one variable i.e f(x_1, x_2, ..., x_n) = x_i for some i in [n]. In this work, we study a k-query dictatorship test. Dictatorship tests are central in proving many hardness results for constraint satisfaction problems. The dictatorship test is said to have perfect completeness if it accepts any dictator function. The soundness of a test is the maximum probability with which it accepts any function far from a dictator. Our main result is a k-query dictatorship test with perfect completeness and soundness (2k + 1)/(2^k), where k is of the form 2^t -1 for any integer t > 2. This improves upon the result of [Tamaki-Yoshida, Random Structures & Algorithms, 2015] which gave a dictatorship test with soundness (2k + 3)/(2^k).