Searcharxiv⌕ Search

arXiv subjects

Edward Wang

Publications and source records attributed to Edward Wang.

At least 19 recordsLinked to original sources
arXiv2026

Embedded Arena: Iterative Optimization via Hardware Feedback

Embedded devices from wildlife monitoring stations to clinical wearables require local AI inference due to latency, communication, or privacy constraints. Optimizing models for heterogeneous microcontrollers (MCUs) requires simultaneously satisfying hard physical constraints on memory, power, and temperature while preserving accuracy, a multidimensional optimization that is today performed manually by experts. We ask whether an LLM agent can autonomously navigate this complex, multi-turn pipeline guided by real hardware feedback, and introduce a hardware-in-the-loop agent arena in which the agent iteratively refines both model and firmware -- compiling, flashing, and measuring on real hardware -- to enable closed-loop optimization. Frontier models, including Claude Opus 4.7 and Gemini 3.1 Pro, fail entirely without hardware feedback (0% deployment success), whereas our hardware-in-the-loop formulation achieves the first successful deployment within three iterations and can surpass human expert results within seven. This agentic co-optimization achieves 250x compression for vision models with <3.3% accuracy loss and 400x for audio with <6% Feature Error Rate loss, enabling battery-free operation on a commercial MCU via solar harvesting. We demonstrate practical impact in two real-world systems: an elk-detection camera trap (96.7% accuracy) and a phonetic-transcription wearable (8.44% FER) for child development research.

cs.AR↗
arXiv2026

Self-consistent computation of pair production from non-relativistic effective field theories in the Keldysh-Schwinger formalism

Sommerfeld-enhanced annihilation cross sections in the presence of nearly zero-energy bound states can become so large that perturbative partial-wave unitarity appears to be violated. Previous literature incorporated the short-distance annihilation potential self-consistently into the computation of the Schr\"odinger wave function at the origin, leading to the unitarization of the Sommerfeld effect in vacuum. We employ non-relativistic effective field theory methods and the Keldysh-Schwinger formalism to additionally include pair-creation effects in the self-consistent computation of four-point correlation functions, which renders the unitarization temperature dependent. Up to small thermal corrections in the non-relativistic and dilute regime of the pairs, we confirm the previous results based on the Schr\"odinger equation approach for scattering states in vacuum. For the first time, we analyze bound-state contributions beyond their leading decay via annihilation. Interestingly, our self-consistent computation of the four-point correlation function shows that bound states remain on-shell in their out-of-equilibrium decay, even though their spectral functions take the form of Breit-Wigner distributions due to finite decay widths. While this may appear paradoxical, it aligns with expectations from earlier results based on exact analytic solutions of the Kadanoff-Baym equations for a decaying elementary particle in a thermal environment.

hep-ph↗
arXiv2026

DECKBench: Benchmarking Multi-Agent Frameworks for Academic Slide Generation and Editing

Automatically generating and iteratively editing academic slide decks requires more than document summarization. It demands faithful content selection, coherent slide organization, layout-aware rendering, and robust multi-turn instruction following. However, existing benchmarks and evaluation protocols do not adequately measure these challenges. To address this gap, we introduce the Deck Edits and Compliance Kit Benchmark (DECKBench), an evaluation framework for multi-agent slide generation and editing. DECKBench is built on a curated dataset of paper to slide pairs augmented with realistic, simulated editing instructions. Our evaluation protocol systematically assesses slide-level and deck-level fidelity, coherence, layout quality, and multi-turn instruction following. We further implement a modular multi-agent baseline system that decomposes the slide generation and editing task into paper parsing and summarization, slide planning, HTML creation, and iterative editing. Experimental results demonstrate that the proposed benchmark highlights strengths, exposes failure modes, and provides actionable insights for improving multi-agent slide generation and editing systems. Overall, this work establishes a standardized foundation for reproducible and comparable evaluation of academic presentation generation and editing. Code and data are publicly available at https://github.com/morgan-heisler/DeckBench .

cs.AI↗
arXiv2026

Straight-line optimality in Bellman's lost-in-a-forest problem for Euclidean balls

We prove that among all unit-speed paths, a straight line minimises the expected escape time from a ball in $\mathbf{R}^n$, solving the min-mean variant of Bellman's Lost~in~a~Forest problem for ball-shaped forests. The proof uses the Kneser--Poulsen conjecture in the plane, together with results on polygonal chain straightening in higher dimensions. Moreover, we calculate this minimal escape time by deriving the expected linear distance to the boundary of a ball in $n$ dimensions.

math.PR↗
arXiv2025

Baryogenesis and EDMs in the 2HDM+CS

We perform the first joint analysis of baryogenesis from initial Higgs charges (also called Higgsogenesis) and EDMs in a model with two Higgs doublets, a complex scalar and a Majorana fermion. In our proposed scenario, baryogenesis happens in three steps: (1) The decay of the scalar produces an asymmetry between the two Higgs doublets which is partly transferred to fermions via Standard Model processes. (2) This asymmetry is converted into a $B-L$ charge via interactions mediated by the Majorana fermion. (3) The weak sphaleron processes partially convert the resulting $B-L$ charge into a $B$ asymmetry. We perform a numerical analysis of baryogenesis and the EDM contributions and find that, due to resonant enhancement, baryogenesis is possible with singlet masses as low as $10^5$ GeV. We also find that for some parameters, the model can give large contributions to the electron and neutron EDM while also producing baryogenesis. The present scenario offers a new perspective on interpreting observational bounds in terms of cosmology: Because it relies on out-of-equilibrium decays, the questions of baryogenesis and $CP$ violation in the Higgs sector may be linked with EDM signals without the additional assumption of a first order electroweak phase transition.

hep-ph↗
arXiv2025

Pick-up Sticks and the Fibonacci Factorial

We present a variation of the broken stick problem in which $n$ stick lengths are sampled uniformly at random. We prove that the probability that no three sticks can form a triangle is the reciprocal of the product of the first $n$ Fibonacci numbers. Extensions to quadrilaterals and general $k$-gons are also discussed.

math.PR↗
arXiv2025

Evaluation of the phi-3-mini SLM for identification of texts related to medicine, health, and sports injuries

Small Language Models (SLMs) have potential to be used for automatically labelling and identifying aspects of text data for medicine/health-related purposes from documents and the web. As their resource requirements are significantly lower than Large Language Models (LLMs), these can be deployed potentially on more types of devices. SLMs often are benchmarked on health/medicine-related tasks, such as MedQA, although performance on these can vary especially depending on the size of the model in terms of number of parameters. Furthermore, these test results may not necessarily reflect real-world performance regarding the automatic labelling or identification of texts in documents and the web. As a result, we compared topic-relatedness scores from Microsofts phi-3-mini-4k-instruct SLM to the topic-relatedness scores from 7 human evaluators on 1144 samples of medical/health-related texts and 1117 samples of sports injury-related texts. These texts were from a larger dataset of about 9 million news headlines, each of which were processed and assigned scores by phi-3-mini-4k-instruct. Our sample was selected (filtered) based on 1 (low filtering) or more (high filtering) Boolean conditions on the phi-3 SLM scores. We found low-moderate significant correlations between the scores from the SLM and human evaluators for sports injury texts with low filtering (\r{ho} = 0.3413, p < 0.001) and medicine/health texts with high filtering (\r{ho} = 0.3854, p < 0.001), and low significant correlation for medicine/health texts with low filtering (\r{ho} = 0.2255, p < 0.001). There was negligible, insignificant correlation for sports injury-related texts with high filtering (\r{ho} = 0.0318, p = 0.4466).

cs.IR↗
arXiv2024

The Neutrino Mass Bound from Leptogenesis Revisited

Recent years have seen a great improvement in the computation of $CP$-conserving and $CP$-violating equilibration rates for leptogenesis. These are relevant for the relativistic regime of the sterile Majorana fermions and the dynamics of the Standard Model particles acting as spectator processes. In order to probe the regime of large $({\cal O}(10^2))$ washout parameters, we add $\Delta L = 2$ washout processes, which we derive in the CTP-formalism. To demonstrate their significance, we apply state-of-the-art computational techniques to a simple yet well-motivated phenomenological scenario: unflavored leptogenesis in a hierarchical type-I seesaw model. We then perform a parameter scan of the final baryon asymmetry and find a constraint $m_\text{lightest} \lesssim 0.15 \, \text{eV}$ on the absolute neutrino mass scale, which is slightly less stringent than previously reported bounds obtained without the aforementioned improvements. The relaxation of the bounds is mainly due to partially equilibrated spectator fields, which protect part of the asymmetry from washout and lead to larger final asymmetries. While this might seem like a minor correction, the actual dynamics of the fields is substantially altered by these effects. Even though we focused on a particularly simple scenario for leptogenesis, the methods employed here can and should be extended to other models, thus giving us a more accurate picture of the different leptogenesis scenarios.

hep-ph↗
arXiv2024

Zero-sum Dynkin games under common and independent Poisson constraints

Zero-sum Dynkin games under Poisson constraints, where players can only stop at the event times of a Poisson process, have been studied widely in the recent literature. The constraint can be modelled in two ways: either both players share the same Poisson process (the common constraint) or each player has their own Poisson process (the independent constraint). In a Markovian framework, where payoffs are functions of an underlying diffusion, we establish necessary and sufficient conditions for the equivalence of the game's solution--comprising the value function and optimal stopping sets--under the common and independent constraints. Specifically, if the stopping sets of the maximiser and minimiser in the game under the common constraint are disjoint, then the solution to the game is the same under both the common and the independent constraint. However, the fact that the stopping sets are disjoint in the game under the independent constraint is not sufficient to guarantee that the solution of the game under the independent constraint is also the solution under the common constraint. To demonstrate the broad applicability of our results, we solve infinite-horizon Dynkin games satisfying the assumptions of our main theorems, using backward stochastic differential equation (BSDE) techniques. This requires extending standard BSDE results from the finite-horizon setting to the infinite-horizon case, allowing for unbounded solutions.

math.OC↗
arXiv2024

Latent Spaces Enable Transformer-Based Dose Prediction in Complex Radiotherapy Plans

Evidence is accumulating in favour of using stereotactic ablative body radiotherapy (SABR) to treat multiple cancer lesions in the lung. Multi-lesion lung SABR plans are complex and require significant resources to create. In this work, we propose a novel two-stage latent transformer framework (LDFormer) for dose prediction of lung SABR plans with varying numbers of lesions. In the first stage, patient anatomical information and the dose distribution are encoded into a latent space. In the second stage, a transformer learns to predict the dose latent from the anatomical latents. Causal attention is modified to adapt to different numbers of lesions. LDFormer outperforms a state-of-the-art generative adversarial network on dose conformality in and around lesions, and the performance gap widens when considering overlapping lesions. LDFormer generates predictions of 3-D dose distributions in under 30s on consumer hardware, and has the potential to assist physicians with clinical decision making, reduce resource costs, and accelerate treatment planning.

physics.med-ph↗
arXiv2024

A Scotogenic Model as a Prototype for Leptogenesis with One Single Gauge Singlet

We investigate the potential of a minimal Scotogenic model with two additional scalar doublets and a single heavy Majorana fermion to explain neutrino masses, dark matter, and the baryon asymmetry of the Universe. In this minimal setup, Leptogenesis is purely flavored, and a second Majorana neutrino is not necessary because the Yukawa couplings of the extra doublets yield the necessary $CP$-odd phases. The mechanism we employ can also be applied to a wide range of scenarios with at least one singlet and two gauge multiplets. Despite stringent limits from the dark matter abundance, direct detection experiments, and the baryon asymmetry of the Universe, we find a parametric region consistent with all bounds which could resolve the above shortcomings of the Standard Model of particle physics. Methodically, we improve on the calculation of correlations between the mixing scalar fields given their finite width. We also present an argument to justify the kinetic equilibrium approximation for out-of-equilibrium distribution functions often used in calculations of Baryogenesis and Leptogenesis.

hep-ph↗
arXiv2023

Resonant Spin-Flavor Precession of Sterile Neutrinos

We analyze the impact of resonant conversions mediated by non-vanishing magnetic moments between active neutrinos and a heavy sterile neutrino on the supernova neutrino flux. We present the level-crossing scheme for such a scenario and derive the neutrino fluxes after conversion, paying particular attention to the order in which the resonances occur. We then compute the expected event rates from the neutronization burst of a future supernova at DUNE and Hyper-Kamiokande to derive new constraints on the neutrino magnetic moment. With this, we find a sensitivity down to a few $10^{-15} \mu_B$ for a sterile neutrino in the $O(\rm{eV})$ mass range.

hep-ph↗
arXiv2023

SpecTracle: Wearable Facial Motion Tracking from Unobtrusive Peripheral Cameras

Facial motion tracking in head-mounted displays (HMD) has the potential to enable immersive "face-to-face" interaction in a virtual environment. However, current works on facial tracking are not suitable for unobtrusive augmented reality (AR) glasses or do not have the ability to track arbitrary facial movements. In this work, we demonstrate a novel system called SpecTracle that tracks a user's facial motions using two wide-angle cameras mounted right next to the visor of a Hololens. Avoiding the usage of cameras extended in front of the face, our system greatly improves the feasibility to integrate full-face tracking into a low-profile form factor. We also demonstrate that a neural network-based model processing the wide-angle cameras can run in real-time at 24 frames per second (fps) on a mobile GPU and track independent facial movement for different parts of the face with a user-independent model. Using a short personalized calibration, the system improves its tracking performance by 42.3% compared to the user-independent model.

cs.HC↗
arXiv2023

AIROGS: Artificial Intelligence for RObust Glaucoma Screening Challenge

The early detection of glaucoma is essential in preventing visual impairment. Artificial intelligence (AI) can be used to analyze color fundus photographs (CFPs) in a cost-effective manner, making glaucoma screening more accessible. While AI models for glaucoma screening from CFPs have shown promising results in laboratory settings, their performance decreases significantly in real-world scenarios due to the presence of out-of-distribution and low-quality images. To address this issue, we propose the Artificial Intelligence for Robust Glaucoma Screening (AIROGS) challenge. This challenge includes a large dataset of around 113,000 images from about 60,000 patients and 500 different screening centers, and encourages the development of algorithms that are robust to ungradable and unexpected input data. We evaluated solutions from 14 teams in this paper, and found that the best teams performed similarly to a set of 20 expert ophthalmologists and optometrists. The highest-scoring team achieved an area under the receiver operating characteristic curve of 0.99 (95% CI: 0.98-0.99) for detecting ungradable images on-the-fly. Additionally, many of the algorithms showed robust performance when tested on three other publicly available datasets. These results demonstrate the feasibility of robust AI-enabled glaucoma screening.

eess.IV↗
arXiv2023

Neuromorphic High-Frequency 3D Dancing Pose Estimation in Dynamic Environment

As a beloved sport worldwide, dancing is getting integrated into traditional and virtual reality-based gaming platforms nowadays. It opens up new opportunities in the technology-mediated dancing space. These platforms primarily rely on passive and continuous human pose estimation as an input capture mechanism. Existing solutions are mainly based on RGB or RGB-Depth cameras for dance games. The former suffers in low-lighting conditions due to the motion blur and low sensitivity, while the latter is too power-hungry, has a low frame rate, and has limited working distance. With ultra-low latency, energy efficiency, and wide dynamic range characteristics, the event camera is a promising solution to overcome these shortcomings. We propose YeLan, an event camera-based 3-dimensional high-frequency human pose estimation(HPE) system that survives low-lighting conditions and dynamic backgrounds. We collected the world's first event camera dance dataset and developed a fully customizable motion-to-event physics-aware simulator. YeLan outperforms the baseline models in these challenging conditions and demonstrated robustness against different types of clothing, background motion, viewing angle, occlusion, and lighting fluctuations.

cs.CV↗
arXiv2022

Magnetic Moments of Astrophysical Neutrinos

We study the impact of neutrino magnetic moments on astrophysical neutrinos, in particular supernova neutrinos and ultra-high energy neutrinos from extragalactic sources. We show that magnetic moment-induced conversion of left-handed neutrinos into unobservable right-handed singlet states can substantially change the flux and flavour composition of these neutrinos at Earth. Notably, neutrinos from a supernova's neutronisation burst, whose flux can be predicted with O(10%) accuracy, offer a discovery reach to neutrino magnetic moments $\sim \text{few} \times 10^{-13} \mu_B$, up to one order of magnitude below current limits. For high-energy neutrinos from distant sources, for which no robust flux prediction exists, we show how the flavour composition at Earth can be used as a handle to establish the presence of non-negligible magnetic moments, potentially down to $\text{few} \times 10^{-17} \mu_B$ if the measurement can be performed on neutrinos from a single source. In both cases, the sensitivity strongly depends on the galactic resp. intergalactic magnetic field profiles along the line of sight. Therefore, while a discovery is possible down to very small values of the magnetic moment, the absence of a discovery does not imply an equally strong limit. We also comment on the dependence of our results on the right-handed neutrino mass, paying special attention to the transition from coherent deflection by a classical magnetic field to incoherent scattering on individual scattering targets. Finally, we show that a measurement of Standard Model Dirac neutrino magnetic moments, of order $10^{-19} \mu_B$, could be possible under rather optimistic, but not completely outrageous, assumptions using flavour ratios of high-energy astrophysical neutrinos.

hep-ph↗
arXiv2021

Callable convertible bonds under liquidity constraints and hybrid priorities

This paper investigates the callable convertible bond problem in the presence of a liquidity constraint modelled by Poisson signals. We assume that neither the bondholder nor the firm has absolute priority when they stop the game simultaneously, but instead, a proportion $m\in[0,1]$ of the bond is converted to the firm's stock and the rest is called by the firm. The paper thus generalizes the special case studied in [Liang and Sun, Dynkin games with Poisson random intervention times, SIAM Journal on Control and Optimization, 57 (2019), 2962-2991] where the bondholder has priority ($m=1$), and presents a complete solution to the callable convertible bond problem with liquidity constraint. The callable convertible bond is an example of a Dynkin game, but falls outside the standard paradigm since the payoffs do not depend in an ordered way upon which agent stops the game. We show how to deal with this non-ordered situation by introducing a new technique which may be of interest in its own right, and then apply it to the bond problem.

q-fin.MF↗