Search for: All records

Deep neural networks are often seen as different from other model classes by defying conventional notions of generalization. Popular examples of anomalous generalization behaviour include benign overfitting, double descent, and the success of overparametrization. We argue that these phenomena are not distinct to neural networks, or particularly mysterious. Moreover, this generalization behaviour can be intuitively understood, and rigorously characterized using long-standing generalization frameworks such as PAC-Bayes and countable hypothesis bounds. We present soft inductive biases as a key unifying principle in explaining these phenomena: rather than restricting the hypothesis space to avoid overfitting, embrace a flexible hypothesis space, with a soft preference for simpler solutions that are consistent with the data. This principle can be encoded in many model classes, and thus deep learning is not as mysterious or different from other model classes as it might seem. However, we also highlight how deep learning is relatively distinct in other ways, such as its ability for representation learning, phenomena such as mode connectivity, and its relative universality. 

The core component of attention is the scoring function, which transforms the inputs into low-dimensional queries and keys and takes the dot product of each pair. While the low-dimensional projection improves efficiency, it causes information loss for certain tasks that have intrinsically high-dimensional inputs. Additionally, attention uses the same scoring function for all input pairs, without imposing a distance-dependent compute bias for neighboring tokens in the sequence. In this work, we address these shortcomings by proposing new scoring functions based on computationally efficient structured matrices with high ranks, including Block Tensor-Train (BTT) and Multi-Level Low Rank (MLR) matrices. On in-context regression tasks with high-dimensional inputs, our proposed scoring functions outperform standard attention for any fixed compute budget. On language modeling, a task that exhibits locality patterns, our MLR-based attention method achieves improved scaling laws compared to both standard attention and variants of sliding window attention. Additionally, we show that both BTT and MLR fall under a broader family of efficient structured matrices capable of encoding either full-rank or distance-dependent compute biases, thereby addressing significant shortcomings of standard attention. Finally, we show that MLR attention has promising results for long-range time-series forecasting. 

Data include soil and litter measurements for moisture, pH, and carbon-to-nitrogen ratio. Samples were collected from 8 different ecoregions, as determined by NEON, at various NEON/LTER and/or other experimental sites. Soil cores and litter samples were taken in the spring and fall of 2022. 

To build effective therapeutics, biologists iteratively mutate antibody sequences to improve binding and stability. Proposed mutations can be informed by previous measurements or by learning from large antibody databases to predict only typical antibodies. Unfortunately, the space of typical antibodies is enormous to search, and experiments often fail to find suitable antibodies on a budget. We introduce Clone-informed Bayesian Optimization (CloneBO), a Bayesian optimization procedure that efficiently optimizes antibodies in the lab by teaching a generative model how our immune system optimizes antibodies. Our immune system makes antibodies by iteratively evolving specific portions of their sequences to bind their target strongly and stably, resulting in a set of related, evolving sequences known as a clonal family. We train a large language model, CloneLM, on hundreds of thousands of clonal families and use it to design sequences with mutations that are most likely to optimize an antibody within the human immune system. We propose to guide our designs to fit previous measurements with a twisted sequential Monte Carlo procedure. We show that CloneBO optimizes antibodies substantially more efficiently than previous methods in realistic in silico experiments and designs stronger and more stable binders in in vitro wet lab experiments. 

Abstract The performance of a caesium fountain frequency reference for use in precision measurements of trapped antihydrogen in the ALPHA experiment at CERN is evaluated. A description of the fountain is provided together with a characterisation of systematic effects. The impact of the magnetic environment in the Antimatter Factory, where the fountain is installed, on the performance of the fountain is considered and shown to be insignificant. The systematic fractional frequency uncertainty of the fountain is 3.0 × 10^-16. The short-term frequency stability of the measured frequency from the ALPHA-HM1 maser is 1.5 × 10^-13τ^-1/2, whereas the fountain itself shows a stability limit of 4.7 × 10^-14τ^-1/2. We find a fractional frequency difference of (1.0 ± 2.2 (stat.) ± 6.5 (syst.)) × 10^-16 in a comparison with Terrestrial Time via a GNSS Common View satellite link between January 2023 and June 2024. The fountain will enables a significant increase in frequency precision in antihydrogen spectroscopic measurements, and paves the way for improved limits on matter-antimatter comparisons. 

FPGAs have been shown to operate reliably within harsh radiation environments by employing single-event upset (SEU) mitigation techniques, such as configuration scrubbing, triple-modular redundancy, error correction coding, and radiation aware implementation techniques. The effectiveness of these techniques, however, is limited when using complex system-level designs that employ complex I/O interfaces with single-point failures. In previous work, a complex SoC system running Linux applied several of these techniques only to obtain an improvement of 14\(\times\)in mean time to failure (MTTF). A detailed post-radiation fault analysis found that the limitations in reliability were due to the DDR interface, the global clock network, and interconnect. This article applied a number of design-specific SEU mitigation techniques to address the limitations in reliability of this design. These changes include triplicating the global clock, optimizing the placement of the reduction output voters and input flip-flops, and employing a mapping technique called “striping.” The application of these techniques improved MTTF of the mitigated design by a factor of 1.54\(\times\)and thus provides a 22.8X\(\times\)MTTF improvement over the unmitigated design. A post-radiation fault analysis using BFAT was also performed to find the remaining design vulnerabilities. 

Bayesian optimization is a highly efficient approach to optimizing objective functions which are expensive to query. These objectives are typically represented by Gaussian process (GP) surrogate models which are easy to optimize and support exact inference. While standard GP surrogates have been well-established in Bayesian optimization, Bayesian neural networks (BNNs) have recently become practical function approximators, with many benefits over standard GPs such as the ability to naturally handle non-stationarity and learn representations for high-dimensional data. In this paper, we study BNNs as alternatives to standard GP surrogates for optimization. We consider a variety of approximate inference procedures for finite-width BNNs, including high-quality Hamiltonian Monte Carlo, low-cost stochastic MCMC, and heuristics such as deep ensembles. We also consider infinite-width BNNs, linearized Laplace approximations, and partially stochastic models such as deep kernel learning. We evaluate this collection of surrogate models on diverse problems with varying dimensionality, number of objectives, non-stationarity, and discrete and continuous inputs. We find: (i) the ranking of methods is highly problem dependent, suggesting the need for tailored inductive biases; (ii) HMC is the most successful approximate inference procedure for fully stochastic BNNs; (iii) full stochasticity may be unnecessary as deep kernel learning is relatively competitive; (iv) deep ensembles perform relatively poorly; (v) infinite-width BNNs are particularly promising, especially in high dimensions.