We introduce Performers, Transformer architectures which can estimate regular (softmax) full-rank-attention Transformers with provable accuracy, but using only linear (as opposed to quadratic) space and time complexity, without relying on any priors such as sparsity or low-rankness. To approximate softmax attentionkernels, Performers use a novel Fast Attention Via positive Orthogonal Random features approach (FAVOR+), which may be of independent interest for scalable kernel methods. FAVOR+ can be also used to efficiently model kernelizable attention mechanisms beyond softmax. This representational power is crucial to accurately compare softmax with other kernels for the first time on large-scale tasks, beyond the reach of regular Transformers, and investigate optimal attention-kernels. Performers are linear architectures fully compatible with regular Transformers and with strong theoretical guarantees: unbiased or nearly-unbiased estimation of the attention matrix, uniform convergence and low estimation variance. We tested Performers on a rich set of tasks stretching from pixel-prediction through text models to protein sequence modeling. We demonstrate competitive results with other examined efficient sparse and dense attention methods, showcasing effectiveness of the novel attention-learning paradigm leveraged by Performers.
In laboratories, mice are housed at 20–24°C, which is below their lower critical temperature (≈30°C). This increased thermal stress has the potential to alter scientific outcomes. Nesting material should allow for improved behavioral thermoregulation and thus alleviate this thermal stress. Nesting behavior should change with temperature and material, and the choice between nesting or thermotaxis (movement in response to temperature) should also depend on the balance of these factors, such that mice titrate nesting material against temperature. Naïve CD-1, BALB/c, and C57BL/6 mice (36 male and 36 female/strain in groups of 3) were housed in a set of 2 connected cages, each maintained at a different temperature using a water bath. One cage in each set was 20°C (Nesting cage; NC) while the other was one of 6 temperatures (Temperature cage; TC: 20, 23, 26, 29, 32, or 35°C). The NC contained one of 6 nesting provisions (0, 2, 4, 6, 8, or 10g), changed daily. Food intake and nest scores were measured in both cages. As the difference in temperature between paired cages increased, feed consumption in NC increased. Nesting provision altered differences in nest scores between the 2 paired temperatures. Nest scores in NC increased with increasing provision. In addition, temperature pairings altered the difference in nest scores with the smallest difference between locations at 26°C and 29°C. Mice transferred material from NC to TC but the likelihood of transfer decreased with increasing provision. Overall, mice of different strains and sexes prefer temperatures between 26–29°C and the shift from thermotaxis to nest building is seen between 6 and 10 g of material. Our results suggest that under normal laboratory temperatures, mice should be provided with no less than 6 grams of nesting material, but up to 10 grams may be needed to alleviate thermal distress under typical temperatures.
Recent genetic and pharmacological studies have suggested that the metabotropic glutamate receptor subtype 5 (mGluR5) may represent a druggable target in identifying new therapeutics for the treatment of various central nervous system disorders including drug abuse. In particular, considerable attention in the mGluR5 field has been devoted to identifying ligands that bind to the allosteric modulatory site, distinct from the site for the primary agonist glutamate. Both 2-methyl-6-(phenylethynyl)pyridine (MPEP) and its analogue 3-[(2-methyl-4-thiazolyl)ethynyl]pyridine (MTEP) have been shown to be selective and potent noncompetitive antagonists of mGluR5. Because of results presented in this study showing that MTEP prevents the reinstatement of cocaine self-administration caused by the presentation of environmental cues previously associated with cocaine availability, we have prepared a series of analogues of MTEP with the aim of gaining a better understanding of the structural features relevant to its antagonist potency and with the ultimate aim of investigating the effects of such compounds in blunting the self-administration of cocaine. These efforts have led to the identification of compounds showing higher potency as mGluR5 antagonists than either MPEP or MTEP. Two compounds 19 and 59 exhibited functional activity as mGluR5 antagonists that are 490 and 230 times, respectively, better than that of MTEP.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.