Recently, semidefinite programming (SDP) techniques have shown great promise in providing accurate Lipschitz bounds for neural networks. Specifically, the LipSDP approach (Fazlyab et al., 2019) has received much attention and provides the least conservative Lipschitz upper bounds that can be computed with polynomial time guarantees. However, one main restriction of LipSDP is that its formulation requires the activation functions to be slope-restricted on $[0,1]$, preventing its further use for more general activation functions such as GroupSort, MaxMin, and Householder. One can rewrite MaxMin activations for example as residual ReLU networks. However, a direct application of LipSDP to the resultant residual ReLU networks is conservative and even fails in recovering the well-known fact that the MaxMin activation is 1-Lipschitz. Our paper bridges this gap and extends LipSDP beyond slope-restricted activation functions. To this end, we provide novel quadratic constraints for GroupSort, MaxMin, and Householder activations via leveraging their underlying properties such as sum preservation. Our proposed analysis is general and provides a unified approach for estimating $\ell_2$ and $\ell_\infty$ Lipschitz bounds for a rich class of neural network architectures, including non-residual and residual neural networks and implicit models, with GroupSort, MaxMin, and Householder activations. Finally, we illustrate the utility of our approach with a variety of experiments and show that our proposed SDPs generate less conservative Lipschitz bounds in comparison to existing approaches.

A crucial property for achieving secure, trustworthy and interpretable deep learning systems is their robustness: small changes to a system's inputs should not result in large changes to its outputs. Mathematically, this means one strives for networks with a small Lipschitz constant. Several recent works have focused on how to construct such Lipschitz networks, typically by imposing constraints on the weight matrices. In this work, we study an orthogonal aspect, namely the role of the activation function. We show that commonly used activation functions, such as MaxMin, as well as all piece-wise linear ones with two segments unnecessarily restrict the class of representable functions, even in the simplest one-dimensional setting. We furthermore introduce the new N-activation function that is provably more expressive than currently popular activation functions. We provide code at https://github.com/berndprach/NActivation.

For general-sum, n-player, strategic games with transferable utility, the Harsanyi-Shapley value provides a computable method to both 1) quantify the strategic value of a player; and 2) make cooperation rational through side payments. We give a simple formula to compute the HS value in normal-form games. Next, we provide two methods to generalize the HS values to stochastic (or Markov) games, and show that one of them may be computed using generalized Q-learning algorithms. Finally, an empirical validation is performed on stochastic grid-games with three or more players. Source code is provided to compute HS values for both the normal-form and stochastic game setting.

Topological data analysis (TDA) is an expanding field that leverages principles and tools from algebraic topology to quantify structural features of data sets or transform them into more manageable forms. As its theoretical foundations have been developed, TDA has shown promise in extracting useful information from high-dimensional, noisy, and complex data such as those used in biomedicine. To improve efficiency, these techniques may employ landmark samplers. The heuristic maxmin procedure obtains a roughly even distribution of sample points by implicitly constructing a cover comprising sets of uniform radius. However, issues arise with data that vary in density or include points with multiplicities, as are common in biomedicine. We propose an analogous procedure, "lastfirst" based on ranked distances, which implies a cover comprising sets of uniform cardinality. We first rigorously define the procedure and prove that it obtains landmarks with desired properties. We then perform benchmark tests and compare its performance to that of maxmin, on feature detection and class prediction tasks involving simulated and real-world biomedical data. Lastfirst is more general than maxmin in that it can be applied to any data on which arbitrary (and not necessarily symmetric) pairwise distances can be computed. Lastfirst is more computationally costly, but our implementation scales at the same rate as maxmin. We find that lastfirst achieves comparable performance on prediction tasks and outperforms maxmin on homology detection tasks. Where the numerical values of similarity measures are not meaningful, as in many biomedical contexts, lastfirst sampling may also improve interpretability.

We present a meta-method for initializing (seeding) the $k$-means clustering algorithm called PNN-smoothing. It consists in splitting a given dataset into $J$ random subsets, clustering each of them individually, and merging the resulting clusterings with the pairwise-nearest-neighbor (PNN) method. It is a meta-method in the sense that when clustering the individual subsets any seeding algorithm can be used. If the computational complexity of that seeding algorithm is linear in the size of the data $N$ and the number of clusters $k$, PNN-smoothing is also almost linear with an appropriate choice of $J$, and in fact only at most a few percent slower in most cases in practice. We show empirically, using several existing seeding methods and testing on several synthetic and real datasets, that this procedure results in systematically better costs. It can even be applied recursively, and easily parallelized. Our implementation is publicly available at https://github.com/carlobaldassi/KMeansPNNSmoothing.jl