arXiv subjects

Joseph Wilson

Publications and source records attributed to Joseph Wilson.

8 recordsLinked to original sources

arXiv2026

Is the Last Layer Sufficient for Uncertainty Quantification?

Epistemic uncertainty quantification (UQ) for deep neural networks (DNNs) is a requirement for safe adoption of AI in mission-critical settings. Several leading methods for UQ linearize DNNs to form Bayesian Generalized Linear Models (GLMs), where epistemic uncertainty is modeled via the predictive posterior distribution. Linearizing around the parameters of the final connected layer of a DNN is a commonly used approximation for reducing the computational burden of such GLMs, though it is often believed to come at the cost of degraded performance. In this work, we compare GLMs arising from full-network and last-layer linearization using both theoretical and empirical approaches. We first employ tools from random matrix theory to conduct a theoretical comparison; this analysis reveals no meaningful improvement in the UQ capabilities of full linearization. Coupled with a large-scale empirical evaluation across a range of modern machine learning tasks, we arrive at the following conclusion: a last-layer approximation yields comparable UQ performance while offering substantially improved computational efficiency.

stat.ML↗

arXiv2026

From Theory to Practice: Code Generation Using LLMs for CAPEC and CWE Frameworks

The increasing complexity and volume of software systems have heightened the importance of identifying and mitigating security vulnerabilities. The existing software vulnerability datasets frequently fall short in providing comprehensive, detailed code snippets explicitly linked to specific vulnerability descriptions, reducing their utility for advanced research and hindering efforts to develop a deeper understanding of security vulnerabilities. To address this challenge, we present a novel dataset that provides examples of vulnerable code snippets corresponding to Common Attack Pattern Enumerations and Classifications (CAPEC) and Common Weakness Enumeration (CWE) descriptions. By employing the capabilities of Generative Pre-trained Transformer (GPT) models, we have developed a robust methodology for generating these examples. Our approach utilizes GPT-4o, Llama and Claude models to generate code snippets that exhibit specific vulnerabilities as described in CAPEC and CWE documentation. This dataset not only enhances the understanding of security vulnerabilities in code but also serves as a valuable resource for training machine learning models focused on automatic vulnerability detection and remediation. Preliminary evaluations suggest that the dataset generated by Large Language Models demonstrates high accuracy and can serve as a reliable reference for vulnerability identification systems. We found consistent results across the three models, with 0.98 cosine similarity among codes. The final dataset comprises 615 CAPEC code snippets in three programming languages: Java, Python, and JavaScript, making it one of the most extensive and diverse resources in this domain.

cs.CR↗

arXiv2025

Uncertainty Quantification with the Empirical Neural Tangent Kernel

While neural networks have demonstrated impressive performance across various tasks, accurately quantifying uncertainty in their predictions is essential to ensure their trustworthiness and enable widespread adoption in critical systems. Several Bayesian uncertainty quantification (UQ) methods exist that are either cheap or reliable, but not both. We propose a post-hoc, sampling-based UQ method for over-parameterized networks at the end of training. Our approach constructs efficient and meaningful deep ensembles by employing a (stochastic) gradient-descent sampling process on appropriately linearized networks. We demonstrate that our method effectively approximates the posterior of a Gaussian process using the empirical Neural Tangent Kernel. Through a series of numerical experiments, we show that our method not only outperforms competing approaches in computational efficiency-often reducing costs by multiple factors-but also maintains state-of-the-art performance across a variety of UQ metrics for both regression and classification tasks.

stat.ML↗

arXiv2024

Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning

Datasets are foundational to many breakthroughs in modern artificial intelligence. Many recent achievements in the space of natural language processing (NLP) can be attributed to the finetuning of pre-trained models on a diverse set of tasks that enables a large language model (LLM) to respond to instructions. Instruction fine-tuning (IFT) requires specifically constructed and annotated datasets. However, existing datasets are almost all in the English language. In this work, our primary goal is to bridge the language gap by building a human-curated instruction-following dataset spanning 65 languages. We worked with fluent speakers of languages from around the world to collect natural instances of instructions and completions. Furthermore, we create the most extensive multilingual collection to date, comprising 513 million instances through templating and translating existing datasets across 114 languages. In total, we contribute four key resources: we develop and open-source the Aya Annotation Platform, the Aya Dataset, the Aya Collection, and the Aya Evaluation Suite. The Aya initiative also serves as a valuable case study in participatory research, involving collaborators from 119 countries. We see this as a valuable framework for future research collaborations that aim to bridge gaps in resources.

cs.CL↗

arXiv2021

Explicit Baker-Campbell-Hausdorff-Dynkin formula for Spacetime via Geometric Algebra

We present a compact Baker-Campbell-Hausdorff-Dynkin formula for the composition of Lorentz transformations $e^{\sigma_i}$ in the spin representation (a.k.a. Lorentz rotors) in terms of their generators $\sigma_i$: $$ \ln(e^{\sigma_1}e^{\sigma_2}) = \tanh^{-1}\left(\frac{ \tanh \sigma_1 + \tanh \sigma_2 + \frac12[\tanh \sigma_1, \tanh \sigma_2] }{ 1 + \frac12\{\tanh \sigma_1, \tanh \sigma_2\} }\right) $$ This formula is general to geometric algebras (a.k.a. real Clifford algebras) of dimension $\leq 4$, naturally generalising Rodrigues' formula for rotations in $\mathbb{R}^3$. In particular, it applies to Lorentz rotors within the framework of Hestenes' spacetime algebra, and provides an efficient method for composing Lorentz generators. Computer implementations are possible with a complex $2\times2$ matrix representation realised by the Pauli spin matrices. The formula is applied to the composition of relativistic $3$-velocities yielding simple expressions for the resulting boost and the concomitant Wigner angle.

math-ph↗

arXiv2020

Optical Convolutional Neural Networks -- Combining Silicon Photonics and Fourier Optics for Computer Vision

The Convolutional Neural Network (CNN) is a state-of-the-art architecture for a wide range of deep learning problems, the quintessential example of which is computer vision. CNNs principally employ the convolution operation, which can be accelerated using the Fourier transform. In this paper, we present an optical hardware accelerator that combines silicon photonics and free-space optics, leveraging the use of the optical Fourier transform within several CNN architectures. The hardware presented is a proof of concept, demonstrating that this technology can be applied to artificial intelligence problems with a large efficiency boost with respect to canonical methods.

eess.IV↗

arXiv2019

Practical Hidden Voice Attacks against Speech and Speaker Recognition Systems

Voice Processing Systems (VPSes), now widely deployed, have been made significantly more accurate through the application of recent advances in machine learning. However, adversarial machine learning has similarly advanced and has been used to demonstrate that VPSes are vulnerable to the injection of hidden commands - audio obscured by noise that is correctly recognized by a VPS but not by human beings. Such attacks, though, are often highly dependent on white-box knowledge of a specific machine learning model and limited to specific microphones and speakers, making their use across different acoustic hardware platforms (and thus their practicality) limited. In this paper, we break these dependencies and make hidden command attacks more practical through model-agnostic (blackbox) attacks, which exploit knowledge of the signal processing algorithms commonly used by VPSes to generate the data fed into machine learning systems. Specifically, we exploit the fact that multiple source audio samples have similar feature vectors when transformed by acoustic feature extraction algorithms (e.g., FFTs). We develop four classes of perturbations that create unintelligible audio and test them against 12 machine learning models, including 7 proprietary models (e.g., Google Speech API, Bing Speech API, IBM Speech API, Azure Speaker API, etc), and demonstrate successful attacks against all targets. Moreover, we successfully use our maliciously generated audio samples in multiple hardware configurations, demonstrating effectiveness across both models and real systems. In so doing, we demonstrate that domain-specific knowledge of audio signal processing represents a practical means of generating successful hidden voice command attacks.

cs.CR↗

arXiv2016

A comprehensive theory of cascading via-paths and the reciprocal pointer chain method

In this paper, we consolidate and expand upon the current theory and potential applications of the set of $k$ best \emph{cascading via-paths} (CVPs) and the \emph{reciprocal pointer chain} (RPC) method for identifying them. CVPs are a collection of up to $|V|$ paths between a source and a target node in a graph $G = (V,E)$, computed using two shortest path trees, that have distinctive properties relative to other path sets. They have been shown to be particularly useful in geospatial applications, where they are an intuitive and efficient means for identifying a set of spatially diverse alternatives to the single shortest path between the source and target. However, spatial diversity is not intrinsic to paths in a graph, and little theory has been developed outside of application to describe the nature of these paths and the RPC method in general. Here we divorce the RPC method from its typical geospatial applications and develop a comprehensive theory of CVPs from an abstract graph-theoretic perspective. Restricting ourselves to properties of the CVPs and of the entire set of $k$-best CVPs that can be computed in $O(|E| + |V| \log |V|)$, we are able to then propose, among other things, new and efficient approaches to problems such as generating a diverse set of paths and to computing the $k$ shortest loopless paths between two nodes in a graph. We conclude by demonstrating the new theory in practice, first for a typical application of finding alternative routes in road networks and then for a novel application of identifying layer-boundaries in ground-penetrating radar (GPR) data. It is our hope that by generalizing the RPC method, providing a sound theoretical foundation, and demonstrating novel uses, we are able to broaden its perceived applicability and stimulate new research in this area, both applied and theoretical.

cs.DM↗