Answering a key point left open in a recent work of Bongers, Guo, Li and Wick, we provide the following lower bound $$ |b|_{text{BMO}_{gamma}(mathbb{R}^2)}lesssim |[b,H_{gamma}]|_{L^p(mathbb{R}^2)to L^p(mathbb{R}^2)}, $$ where $H_{gamma}$ is the parabolic Hilbert transform.
We find a minimal notion of non-degeneracy for bilinear singular integral operators $T$ and identify testing conditions on the multiplying function $b$ that characterize the $L^ptimes L^qto L^r,$ $1<p,q<infty$ and $r>frac{1}{2},$ boundedness of the b
ilinear commutator $[b,T]_1(f,g) = bT(f,g) - T(bf,g).$ Our arguments cover almost all arrangements of the integrability exponents $p,q,r,$ with a single open problem presented in the end. Additionally, the arguments extend to the multilinear setting.
We study the commutators $[b,T]$ of pointwise multiplications and bi-parameter Calderon-Zygmund operators and characterize their off-diagonal $L^{p_1}L^{p_2} to L^{q_1}L^{q_2}$ boundedness in the range $(1,infty)$ for several of the mixed norm integrability exponents.
Deep neural networks, including reinforcement learning agents, have been proven vulnerable to small adversarial changes in the input, thus making deploying such networks in the real world problematic. In this paper, we propose RADIAL-RL, a method to
train reinforcement learning agents with improved robustness against any $l_p$-bounded adversarial attack. By simply minimizing an upper bound of the loss functions under worst case adversarial perturbation derived from efficient robustness verification methods, we significantly improve robustness of RL-agents trained on Atari-2600 games and show that RADIAL-RL can beat state-of-the-art robust training algorithms when evaluated against PGD-attacks. We also propose a new evaluation method, Greedy Worst-Case Reward (GWC), for measuring attack agnostic robustness of RL agents. GWC can be evaluated efficiently and it serves as a good estimate of the reward under the worst possible sequence of adversarial attacks; in particular, GWC accounts for the importance of each action and their temporal dependency, improving upon previous approaches that only evaluate whether each single action can change under input perturbations. Our code is available at https://github.com/tuomaso/radial_rl.
We present a pair of joint conditions on the two functions $b_1,b_2$ strictly weaker than $b_1,b_2in operatorname{BMO}$ that almost characterize the $L^2$ boundedness of the iterated commutator $[b_2,[b_1,T]]$ of these functions and a Calderon-Zygmun
d operator $T.$ Namely, we sandwich this boundedness between two bisublinear mean oscillation conditions of which one is a slightly bumped up version of the other.
