arXiv subjects

Benjamin Adams

Publications and source records attributed to Benjamin Adams.

17 recordsLinked to original sources

arXiv2026

Data Structures for Private Token Transfers in TEE-Based Networks

Trusted execution environment (TEE) based confidential smart contract networks promise privacy but remain vulnerable to storage access pattern attacks that can link senders and recipients in token transfers. When contracts update recipient balances during transfers, the unique storage keys accessed reveal transaction relationships even when data is encrypted. This paper introduces two novel data structures to address this vulnerability: the Delayed Write Buffer (DWB) and the Bitwise-Trie of Bucketed Entries (BTBE). The DWB delays recipient balance updates by buffering pending transfers and randomly settling entries, breaking the direct correlation between transfer execution and recipient storage access. The BTBE further enhances privacy by grouping addresses into constant-sized buckets, preventing flooding attacks and creating anonymity sets for balance queries. Additionally, we present a private notification system enabling real-time, privacy-preserving push notifications for confidential contracts. Our domain-specific approach leverages the unique characteristics of token transfers -- asymmetric balance updates and tolerance for delayed settlement -- to achieve practical performance with probabilistic anonymity guarantees.

cs.CR↗

arXiv2025

KEA Explain: Explanations of Hallucinations using Graph Kernel Analysis

Large Language Models (LLMs) frequently generate hallucinations: statements that are syntactically plausible but lack factual grounding. This research presents KEA (Kernel-Enriched AI) Explain: a neurosymbolic framework that detects and explains such hallucinations by comparing knowledge graphs constructed from LLM outputs with ground truth data from Wikidata or contextual documents. Using graph kernels and semantic clustering, the method provides explanations for detected hallucinations, ensuring both robustness and interpretability. Our framework achieves competitive accuracy in detecting hallucinations across both open- and closed-domain tasks, and is able to generate contrastive explanations, enhancing transparency. This research advances the reliability of LLMs in high-stakes domains and provides a foundation for future work on precision improvements and multi-source knowledge integration.

cs.LG↗

arXiv2025

Distilled Circuits: A Mechanistic Study of Internal Restructuring in Knowledge Distillation

Knowledge distillation compresses a larger neural model (teacher) into smaller, faster student models by training them to match teacher outputs. However, the internal computational transformations that occur during this process remain poorly understood. We apply techniques from mechanistic interpretability to analyze how internal circuits, representations, and activation patterns differ between teachers and students. Focusing on GPT2 and its distilled counterpart DistilGPT2, and generalizing our findings to both bidirectional architectures and larger model pairs, we find that student models can reorganize, compress, and discard teacher components, often resulting in a stronger reliance on fewer individual components. To quantify functional alignment beyond output similarity, we introduce an alignment metric based on influence-weighted component similarity, validated across multiple tasks. Our findings reveal that while knowledge distillation preserves broad functional behaviors, it also causes significant shifts in internal computation, with important implications for the robustness and generalization capacity of distilled models.

cs.LG↗

arXiv2025

Detecting Linguistic Diversity on Social Media

This chapter explores the efficacy of using social media data to examine changing linguistic behaviour of a place. We focus our investigation on Aotearoa New Zealand where official statistics from the census is the only source of language use data. We use published census data as the ground truth and the social media sub-corpus from the Corpus of Global Language Use as our alternative data source. We use place as the common denominator between the two data sources. We identify the language conditions of each tweet in the social media data set and validated our results with two language identification models. We then compare levels of linguistic diversity at national, regional, and local geographies. The results suggest that social media language data has the possibility to provide a rich source of spatial and temporal insights on the linguistic profile of a place. We show that social media is sensitive to demographic and sociopolitical changes within a language and at low-level regional and local geographies.

cs.CL↗

arXiv2024

Non-Fungible Programs: Private Full-Stack Applications for Web3

The greatest advantage that Web3 applications offer over Web 2.0 is the evolution of the data access layer. Opaque, centralized services that compelled trust from users are replaced by trustless, decentralized systems of smart contracts. However, the public nature of blockchain-based databases, on which smart contracts transact, has typically presented a challenge for applications that depend on data privacy or that rely on participants having incomplete information. This has changed with the introduction of confidential smart contract networks that encrypt the memory state of active contracts as well as their databases stored on-chain. With confidentiality, contracts can more readily implement novel interaction mechanisms that were previously infeasible. Meanwhile, in both Web 2.0 and Web3 applications the user interface continues to play a crucial role in translating user intent into actionable requests. In many cases, developers have shifted intelligence and autonomy into the client-side, leveraging Web technologies for compute, graphics, and networking. Web3's reliance on such frontends has revealed a pain point though, namely that decentralized applications are not accessible to end users without a persistent host serving the application. Here we introduce the Non-Fungible Program (NFP) model for developing self-contained frontend applications that are distributed via blockchain, powered by Web technology, and backed by private databases persisted in encrypted smart contracts. Access to frontend code, as well as backend services, is controlled and guaranteed by smart contracts according to the NFT ownership model, eliminating the need for a separate host. By extension, NFP applications bring interactivity to token owners and enable new functionalities, such as authorization mechanisms for oracles, supplementary Web services, and overlay networks in a secure manner. In addition...

cs.CY↗

arXiv2024

Pre-Trained Language Models Represent Some Geographic Populations Better Than Others

This paper measures the skew in how well two families of LLMs represent diverse geographic populations. A spatial probing task is used with geo-referenced corpora to measure the degree to which pre-trained language models from the OPT and BLOOM series represent diverse populations around the world. Results show that these models perform much better for some populations than others. In particular, populations across the US and the UK are represented quite well while those in South and Southeast Asia are poorly represented. Analysis shows that both families of models largely share the same skew across populations. At the same time, this skew cannot be fully explained by sociolinguistic factors, economic factors, or geographic factors. The basic conclusion from this analysis is that pre-trained models do not equally represent the world's population: there is a strong skew towards specific geographic populations. This finding challenges the idea that a single model can be used for all populations.

cs.CL↗

arXiv2023

Magneto-optical trap performance for high-bandwidth applications

We study the dynamics of a magneto-optical trap (MOT) operating at high-bandwidth. We find the absolute importance of high recapture efficiency between cycles to maintain a practical atom number. We develop a simple model accounting for MOT trapping forces and pressure induced collisions and validate with experimental data using $\mathrm{{}^{87}Rb}$. This is then applied to quantum sensing predicting a shot noise limited sensitivity of $\mathrm{10^{-7}g/\sqrt{Hz}}$ for a gravimeter at 100 Hz operation. The results are useful for understanding MOT operation at high-bandwidth, particularly in the context of developing mobile high-bandwidth quantum inertial sensors targeting dynamic environments and navigation applications.

physics.atom-ph↗

arXiv2023

Comparing Measures of Linguistic Diversity Across Social Media Language Data and Census Data at Subnational Geographic Areas

This paper describes a preliminary study on the comparative linguistic ecology of online spaces (i.e., social media language data) and real-world spaces in Aotearoa New Zealand (i.e., subnational administrative areas). We compare measures of linguistic diversity between these different spaces and discuss how social media users align with real-world populations. The results from the current study suggests that there is potential to use online social media language data to observe spatial and temporal changes in linguistic diversity at subnational geographic areas; however, further work is required to understand how well social media represents real-world behaviour.

cs.CL↗

arXiv2023

cantnlp@LT-EDI-2023: Homophobia/Transphobia Detection in Social Media Comments using Spatio-Temporally Retrained Language Models

This paper describes our multiclass classification system developed as part of the LTEDI@RANLP-2023 shared task. We used a BERT-based language model to detect homophobic and transphobic content in social media comments across five language conditions: English, Spanish, Hindi, Malayalam, and Tamil. We retrained a transformer-based crosslanguage pretrained language model, XLMRoBERTa, with spatially and temporally relevant social media language data. We also retrained a subset of models with simulated script-mixed social media language data with varied performance. We developed the best performing seven-label classification system for Malayalam based on weighted macro averaged F1 score (ranked first out of six) with variable performance for other language and class-label conditions. We found the inclusion of this spatio-temporal data improved the classification performance for all language and task conditions when compared with the baseline. The results suggests that transformer-based language classification systems are sensitive to register-specific and language-specific retraining.

cs.CL↗

arXiv2021

Quoka Atlas of Scholarly Knowledge Production: An Interactive Sensemaking Tool for Exploring the Outputs of Research Institutions

The vast amount of research produced at institutions world-wide is extremely diverse, and coarse-grained quantitative measures of impact often obscure the individual contributions of these institutions to specific research fields and topics. We show that by applying an information retrieval model to index research articles which are faceted by institution and time, we can develop tools to rank institutions given a keyword query. We present an interactive atlas, Quoka, designed to enable a user to explore these rankings contextually by geography and over time. Through a set of use cases we demonstrate that the atlas can be used to perform sensemaking tasks to learn and collect information about the relationships between institutions and scholarly knowledge production.

cs.DL↗

arXiv2020

We Need to Rethink How We Describe and Organize Spatial Information: Instrumenting and Observing the Community of Users to Improve Data Description and Discovery

In Spatial Data Infrastructure or Cyber Infrastructure, the description of geographic data semantics is intended to support data discovery, reuse and integration. In the vast majority of cases the producers of these data generate descriptions based on particular understandings of what uses the data are good for. This producer-oriented perspective means that the descriptions often do not help to answer the question of whether a data set is of use for a consumer who might want to apply it in a different context. In this paper, we discuss the role geographic information observatories can play in providing an infrastructure for observing the context of data use by consumers. These observations of data pragmatics lead to operational statistical methods that will support better fitness-for-use assessment. Finally, we highlight some of the challenges to building these observatories, and briefly discuss strategies to address those challenges.

cs.SI↗

arXiv2020

A country comparison of place-based activity response to COVID-19 policies

The emergence of the novel Coronavirus Disease in late 2019 (COVID-19) and subsequent pandemic led to an immense disruption in the daily lives of almost everyone on the planet. Faced with the consequences of inaction, most national governments responded with policies that restricted the activities conducted by their inhabitants. As schools and businesses shuttered, the mobility of these people decreased. This reduction in mobility, and related activities, was recorded through ubiquitous location-enabled personal mobile devices. Patterns emerged that varied by place-based activity. In this work the differences in these place-based activity patterns are investigated across nations, specifically focusing on the relationship between government enacted policies and changes in community activity patterns. We show that people's activity response to government action varies widely both across nations as well as regionally within them. Three assessment measures are devised and the results correlate with a number of global indices. We discuss these findings and the relationship between government action and residents' response.

cs.SI↗

arXiv2019

Spatial Data Science: Closing the human-spatial computing-environment loop

Over the last decade, the term spatial computing has grown to have two different, though not entirely unrelated, definitions. The first definition of spatial computing stems from industry, where it refers primarily to new kinds of augmented, virtual, mixed-reality, and natural user interface technologies. A second definition coming out of academia takes a broader perspective that includes active research in geographic information science as well as the aforementioned novel UI technologies. Both senses reflect an ongoing shift toward increased interaction with computing interfaces and sensors embedded in the environment and how the use of these technologies influence how we behave and make sense of and even change the world we live in. Regardless of the definition, research in spatial computing is humming along nicely without the need to identify new research agendas or new labels for communities of researchers. However, as a field of research, it could be helpful to view spatial data science as the glue that coheres spatial computing with problem-solving and learning in the real world into a more holistic discipline.

cs.SI↗

arXiv2019

Understanding population fluctuations through volunteered geographic information and novel indicators: The experience of Rakiura, Stewart Island, New Zealand

In an era of heterogeneous data, novel methods and volunteered geographic information provide opportunities to understand how people interact with a place. However, it is not enough to simply have such heterogeneous data, instead an understanding of its usability and reliability needs to be undertaken. Here, we draw upon the case study of Rakiura, Stewart Island where manifested passenger numbers across the Foveaux Strait are known. We have built a population model to ground truth such novel indicators. In our preliminary study, we find that a number of indicators offer the opportunity to understand fluctuations in populations. Some indicators (such as wastewater volumes) can suggest relative changes in populations in a raw form. While other indicators (such as TripAdvisor reviews or Instagram posts) require further data enrichment to get insights into population fluctuations. This research forms part of a larger research project looking to test and apply such novel indicators to inform disaster risk assessments.

cs.SI↗

arXiv2019

The X-ray and radio activity of typical and luminous Lya emitters from z~2 to z~6: evidence for a diverse, evolving population

Despite recent progress in understanding Lyman-alpha (Lya) emitters (LAEs), relatively little is known regarding their typical black hole activity across cosmic time. Here, we study the X-ray and radio properties of ~4000 LAEs at 2.2 3) and find an average luminosity of 10^{44.31+-0.01} erg/s and an average black hole accretion rate (BHAR) of 0.72+-0.01 Msun/yr, consistent with moderate to high accreting AGN. We detect 120 sources in deep radio data (radio AGN fraction of 3.2%+-0.3%). The global AGN fraction (8.6%+-0.4%) rises with Lya luminosity and declines with increasing redshift. For X-ray detected LAEs, Lya luminosities correlate with the BHARs, suggesting that Lya luminosity becomes an accretion rate indicator. Most LAEs (93.1+-0.6%) at 2<z<6 have no detectable X-ray emission (BHARs<0.017 Msun/yr). The median star formation rate (SFR) of star-forming LAEs from Lya and radio luminosities is 7.6^{+6.6}_{-2.8} Msun/yr. The black hole to galaxy growth ratio (BHAR/SFR) for LAEs is <0.0022, consistent with typical star forming galaxies and the local BHAR/SFR relation. We conclude that LAEs at 2<z<6 include two different populations: an AGN population, where Lya luminosity traces BHAR, making them bright in Lya, and another with low SFRs which remain undetected in even the deepest X-ray stacks but is detected in the radio stacks.

astro-ph.GA↗

arXiv2018

Self-Attention: A Better Building Block for Sentiment Analysis Neural Network Classifiers

Sentiment Analysis has seen much progress in the past two decades. For the past few years, neural network approaches, primarily RNNs and CNNs, have been the most successful for this task. Recently, a new category of neural networks, self-attention networks (SANs), have been created which utilizes the attention mechanism as the basic building block. Self-attention networks have been shown to be effective for sequence modeling tasks, while having no recurrence or convolutions. In this work we explore the effectiveness of the SANs for sentiment analysis. We demonstrate that SANs are superior in performance to their RNN and CNN counterparts by comparing their classification accuracy on six datasets as well as their model characteristics such as training speed and memory consumption. Finally, we explore the effects of various SAN modifications such as multi-head attention as well as two methods of incorporating sequence position information into SANs.

cs.CL↗

arXiv2012

The observational roots of reference of the semantic web

Shared reference is an essential aspect of meaning. It is also indispensable for the semantic web, since it enables to weave the global graph, i.e., it allows different users to contribute to an identical referent. For example, an essential kind of referent is a geographic place, to which users may contribute observations. We argue for a human-centric, operational approach towards reference, based on respective human competences. These competences encompass perceptual, cognitive as well as technical ones, and together they allow humans to inter-subjectively refer to a phenomenon in their environment. The technology stack of the semantic web should be extended by such operations. This would allow establishing new kinds of observation-based reference systems that help constrain and integrate the semantic web bottom-up.

cs.AI↗