The names of variables and functions serve as implicit documentation and are instrumental for program comprehension. But choosing good meaningful names is hard. We perform a sequence of experiments in which a total of 334 subjects are required to choose names in given programming scenarios. The first experiment shows that the probability that two developers would select the same name is low: in the 47 instances in our experiments the median probability was only 6.9 percent. At the same time, given that a specific name is chosen, it is usually understood by the majority of developers. Analysis of the names given in the experiment suggests a model where naming is a (not necessarily cognizant or serial) three-step process: (1) selecting the concepts to include in the name, (2) choosing the words to represent each concept, and (3) constructing a name using these words. A followup experiment, using the same experimental setup, then checked whether using this model explicitly can improve the quality of names. The results were that names selected by subjects using the model were judged by two independent judges to be superior to names chosen in the original experiment by a ratio of two-to-one. Using the model appears to encourage the use of more concepts and longer names.

Heavy tailed and contaminated data are common in various applications of machine learning. A standard technique to handle regression tasks that involve such data, is to use robust losses, e.g., the popular Huber’s loss. In structured problems, however, where there are multiple labels and structural constraints on the labels are imposed (or learned), robust optimization is challenging, and more often than not the loss used is simply the negative log-likelihood of a Gaussian Markov random field. Heavy tailed and contaminated data are common in various applications of machine learning. A standard technique to handle regression tasks that involve such data, is to use robust losses, e.g., the popular Huber’s loss. In structured problems, however, where there are multiple labels and structural constraints on the labels are imposed (or learned), robust optimization is challenging, and more often than not the loss used is simply the negative log-likelihood of a Gaussian Markov random field. In this work, we analyze robust alternatives. Theoretical understanding of such problems is quite limited, with guarantees on optimization given only for special cases and non-structured settings. The core of the difficulty is the non-convexity of the objective function, implying that standard optimization algorithms may converge to sub-optimal critical points. Our analysis focuses on loss functions that arise from elliptical distributions, which appealingly include most loss functions proposed in the literature as special cases. We show that, even though these problems are non-convex, they can be optimized efficiently. Concretely, we prove that at the limit of infinite training data, due to algebraic properties of the problem, all stationary points are globally optimal. Finally, we demonstrate the empirical appeal of using these losses for regression on synthetic and real-life data.