arXiv subjects

Ronin Wu

Publications and source records attributed to Ronin Wu.

At least 19 recordsLinked to original sources

arXiv2026

Implementations of Quantum and Classical Topology-Aligned Architectures for Molecular Property Prediction

For low-data and resource-constrained regimes typical of quantum chemistry, parameter-efficient learning is a key objective. Here, we propose a topology-aligned inductive bias in which the model architecture mirrors the molecular bond graph: atoms map to a fixed register of computational units, and bonds determine which pairs interact through shared learnable parameters. This principle is instantiated in two architectures: a variational quantum circuit (Iso-QGNN) and a parameter-matched classical message-passing network (Iso-CGNN). The models are benchmarked on HOMO-LUMO and dipole moment binary classification tasks over the QM9 benchmark. With 64 trainable parameters, the implementations achieve test AUCs of approximately 0.89 (quantum) and 0.92 (classical) on the gap task, and close to 0.78 (both) on the dipole task. The models reach 90% of asymptotic performance within about 300 training molecules and gradient norms remain stable throughout training. These results indicate that the topology-aligned inductive bias is the active ingredient driving parameter efficiency at QM9 scale, with implications for matched-baseline benchmarking in quantum machine learning.

cs.LG↗

arXiv2026

Near- to Far-Infrared Spectral Energy Distribution Analysis of Interacting Galaxies in Hickson Compact Groups 56 and Stephan's Quintet (HCG 92)

We investigate star formation activity in galaxies belonging to two Hickson Compact Groups (HCGs), HCG 56 and HCG 92 (Stephan's Quintet), both of which show clear evidence of interactions, using spectral energy distribution (SED) analysis across the near- to far-infrared range. By combining data from the Infrared Satellite AKARI, the Spitzer Space Telescope, and the Herschel Space Observatory, we examine how galactic interactions influence the physical conditions and the evolution of group members. The observed SEDs of member galaxies are compared with model SEDs representing both star-forming galaxies and active galactic nuclei (AGN). Star formation rates (SFRs) are estimated using two independent methods: (i) the strength of mid-infrared polycyclic aromatic hydrocarbon (PAH) bands and (ii) far-infrared luminosities attributed to star formation, as derived from the models. Although both methods yield generally consistent results, SFRs based on PAH features are systematically lower, possibly due to the PAH destruction in some interacting galaxies. When plotted against the stellar mass, all member galaxies are found below the main sequence of star-forming galaxies in the SDSS field, suggesting that interaction-induced starbursts are not seen in HCG 56 and HCG 92.

astro-ph.GA↗

arXiv2025

Evaluating Embedding Frameworks for Scientific Domain

Finding an optimal word representation algorithm is particularly important in terms of domain specific data, as the same word can have different meanings and hence, different representations depending on the domain and context. While Generative AI and transformer architecture does a great job at generating contextualized embeddings for any given work, they are quite time and compute extensive, especially if we were to pre-train such a model from scratch. In this work, we focus on the scientific domain and finding the optimal word representation algorithm along with the tokenization method that could be used to represent words in the scientific domain. The goal of this research is two fold: 1) finding the optimal word representation and tokenization methods that can be used in downstream scientific domain NLP tasks, and 2) building a comprehensive evaluation suite that could be used to evaluate various word representation and tokenization algorithms (even as new ones are introduced) in the scientific domain. To this end, we build an evaluation suite consisting of several downstream tasks and relevant datasets for each task. Furthermore, we use the constructed evaluation suite to test various word representation and tokenization algorithms.

cs.CL↗

arXiv2025

Fast, Accurate and Interpretable Graph Classification with Topological Kernels

We introduce a novel class of explicit feature maps based on topological indices that represent each graph by a compact feature vector, enabling fast and interpretable graph classification. Using radial basis function kernels on these compact vectors, we define a measure of similarity between graphs. We perform evaluation on standard molecular datasets and observe that classification accuracies based on single topological-index feature vectors underperform compared to state-of-the-art substructure-based kernels. However, we achieve significantly faster Gram matrix evaluation -- up to $20\times$ faster -- compared to the Weisfeiler--Lehman subtree kernel. To enhance performance, we propose two extensions: 1) concatenating multiple topological indices into an \emph{Extended Feature Vector} (EFV), and 2) \emph{Linear Combination of Topological Kernels} (LCTK) by linearly combining Radial Basis Function kernels computed on feature vectors of individual topological graph indices. These extensions deliver up to $12\%$ percent accuracy gains across all the molecular datasets. A complexity analysis highlights the potential for exponential quantum speedup for some of the vector components. Our results indicate that LCTK and EFV offer a favourable trade-off between accuracy and efficiency, making them strong candidates for practical graph learning applications.

cs.LG↗

arXiv2025

QCA-MolGAN: Quantum Circuit Associative Molecular GAN with Multi-Agent Reinforcement Learning

Navigating the vast chemical space of molecular structures to design novel drug molecules with desired target properties remains a central challenge in drug discovery. Recent advances in generative models offer promising solutions. This work presents a novel quantum circuit Born machine (QCBM)-enabled Generative Adversarial Network (GAN), called QCA-MolGAN, for generating drug-like molecules. The QCBM serves as a learnable prior distribution, which is associatively trained to define a latent space aligning with high-level features captured by the GANs discriminator. Additionally, we integrate a novel multi-agent reinforcement learning network to guide molecular generation with desired targeted properties, optimising key metrics such as quantitative estimate of drug-likeness (QED), octanol-water partition coefficient (LogP) and synthetic accessibility (SA) scores in conjunction with one another. Experimental results demonstrate that our approach enhances the property alignment of generated molecules with the multi-agent reinforcement learning agents effectively balancing chemical properties.

quant-ph↗

arXiv2025

Neural Network Assisted Fermionic Compression Encoding: A Lossy-QSCI Framework for Resource-Efficient Ground-State Simulations

Quantum computing promises to revolutionize many-body simulations for quantum chemistry, but its potential is constrained by limited qubits and noise in current devices. In this work, we introduce the Lossy Quantum Selected Configuration Interaction (Lossy-QSCI) framework, which combines a lossy subspace Hamiltonian preparation pipeline with a generic QSCI selection process. This framework integrates a chemistry-inspired lossy Random Linear Encoder (Chemical-RLE) with a neural network-assisted Fermionic Expectation Decoder (NN-FED). The RLE leverages fermionic number conservation to compress quantum states, reducing qubit requirements to O(N log M) for M spin orbitals and N electrons, while preserving crucial ground state information and enabling self-consistent configuration recovery. NN-FED, powered by a neural network trained with minimal data, efficiently decodes these compressed states, overcoming the measurement challenges common in the approaches of the traditional QSCI and its variants. Through iterative quantum sampling and classical post-processing, our hybrid method refines ground state estimates with high efficiency. This framework offers a resource-efficient pathway for ground-state simulations on near-term noisy hardware and could inspire resource-efficient extensions to future devices by minimizing qubit overhead.

quant-ph↗

arXiv2023

Benchmarking and Analyzing In-context Learning, Fine-tuning and Supervised Learning for Biomedical Knowledge Curation: a focused study on chemical entities of biological interest

Automated knowledge curation for biomedical ontologies is key to ensure that they remain comprehensive, high-quality and up-to-date. In the era of foundational language models, this study compares and analyzes three NLP paradigms for curation tasks: in-context learning (ICL), fine-tuning (FT), and supervised learning (ML). Using the Chemical Entities of Biological Interest (ChEBI) database as a model ontology, three curation tasks were devised. For ICL, three prompting strategies were employed with GPT-4, GPT-3.5, BioGPT. PubmedBERT was chosen for the FT paradigm. For ML, six embedding models were utilized for training Random Forest and Long-Short Term Memory models. Five setups were designed to assess ML and FT model performance across different data availability scenarios.Datasets for curation tasks included: task 1 (620,386), task 2 (611,430), and task 3 (617,381), maintaining a 50:50 positive versus negative ratio. For ICL models, GPT-4 achieved best accuracy scores of 0.916, 0.766 and 0.874 for tasks 1-3 respectively. In a direct comparison, ML (trained on ~260,000 triples) outperformed ICL in accuracy across all tasks. (accuracy differences: +.11, +.22 and +.17). Fine-tuned PubmedBERT performed similarly to leading ML models in tasks 1 & 2 (F1 differences: -.014 and +.002), but worse in task 3 (-.048). Simulations revealed performance declines in both ML and FT models with smaller and higher imbalanced training data. where ICL (particularly GPT-4) excelled in tasks 1 & 3. GPT-4 excelled in tasks 1 and 3 with less than 6,000 triples, surpassing ML/FT. ICL underperformed ML/FT in task 2.ICL-augmented foundation models can be good assistants for knowledge curation with correct prompting, however, not making ML and FT paradigms obsolete. The latter two require task-specific data to beat ICL. In such cases, ML relies on small pretrained embeddings, minimizing computational demands.

cs.LG↗

arXiv2022

Searching for Carriers of the Diffuse Interstellar Bands Across Disciplines, using Natural Language Processing

The explosion of scientific publications overloads researchers with information. This is even more dramatic for interdisciplinary studies, where several fields need to be explored. A tool to help researchers overcome this is Natural Language Processing (NLP): a machine-learning (ML) technique that allows scientists to automatically synthesize information from many articles. As a practical example, we have used NLP to conduct an interdisciplinary search for compounds that could be carriers for Diffuse Interstellar Bands (DIBs), a long-standing open question in astrophysics. We have trained a NLP model on a corpus of 1.5 million cross-domain articles in open access, and fine-tuned this model with a corpus of astrophysical publications about DIBs. Our analysis points us toward several molecules, studied primarily in biology, having transitions at the wavelengths of several DIBs and composed of abundant interstellar atoms. Several of these molecules contain chromophores, small molecular groups responsible for the molecule's colour, that could be promising candidate carriers. Identifying viable carriers demonstrates the value of using NLP to tackle open scientific questions, in an interdisciplinary manner.

cs.CL↗

arXiv2022

Leveraging knowledge graphs to update scientific word embeddings using latent semantic imputation

The most interesting words in scientific texts will often be novel or rare. This presents a challenge for scientific word embedding models to determine quality embedding vectors for useful terms that are infrequent or newly emerging. We demonstrate how \gls{lsi} can address this problem by imputing embeddings for domain-specific words from up-to-date knowledge graphs while otherwise preserving the original word embedding model. We use the MeSH knowledge graph to impute embedding vectors for biomedical terminology without retraining and evaluate the resulting embedding model on a domain-specific word-pair similarity task. We show that LSI can produce reliable embedding vectors for rare and OOV terms in the biomedical domain.

cs.CL↗

arXiv2021

Domain-adaptation of spherical embeddings

Domain adaptation of embedding models, updating a generic embedding to the language of a specific domain, is a proven technique for domains that have insufficient data to train an effective model from scratch. Chemistry publications is one such domain, where scientific jargon and overloaded terminology inhibit the performance of a general language model. The recent spherical embedding model (JoSE) proposed in arXiv:1911.01196 jointly learns word and document embeddings during training on the multi-dimensional unit sphere, which performs well for document classification and word correlation tasks. But, we show a non-convergence caused by global rotations during its training prevents it from domain adaptation. In this work, we develop methods to counter the global rotation of the embedding space and propose strategies to update words and documents during domain specific training. Two new document classification data-sets are collated from general and chemistry scientific journals to compare the proposed update training strategies with benchmark models. We show that our strategies are able to reduce the performance cost of domain adaptation to a level similar to Word2Vec.

cs.CL↗

arXiv2019

Investigation of the Origin of the Anomalous Microwave Emission in Lambda Orionis

The anomalous microwave emission (AME) still lacks a conclusive explanation. This excess of emission, roughly between 10 and 50 GHz, tends to defy attempts to explain it as synchrotron or free-free emission. The overlap with frequencies important for cosmic microwave background explorations, combined with a strong correlation with interstellar dust, drive cross-disciplinary collaboration between interstellar medium and observational cosmology. The apparent relationship with dust has prompted a ``spinning dust'' hypothesis. The typical peak frequency range of the AME profile implicates spinning grains on the order of 1 nm. This points to polycyclic aromatic hydrocarbons (PAHs). We use data from the AKARI/Infrared Camera (IRC), due to its thorough PAH-band coverage, to compare AME from the Planck Collaboration astrophysical component separation product with infrared dust emission in the Orionis AME-prominent region. We look also at infrared dust emission from other mid IR and far-IR bands. The results and discussion contained here apply to an angular scale of approximately 1°. We find that certainly dust mass correlates with AME, and that PAH-related emission in the AKARI/IRC 9 μm band correlates slightly more strongly. Using hierarchical Bayesian inference and full dust spectral energy distribution (SED) modeling we argue that AME in λOrionis correlates more strongly with PAH mass than with total dust mass, lending support for a spinning PAH hypothesis within this region. We emphasize that future efforts to understand AME should focus on individual regions, and a detailed comparison of the PAH features with the variation of the AME SED.

astro-ph.GA↗

arXiv2019

Radiative and mechanical feedback into the molecular gas in the Large Magellanic Cloud. II. 30 Doradus

With an aim of probing the physical conditions and excitation mechanisms of warm molecular gas in individual star-forming regions, we performed Herschel SPIRE FTS observations of 30 Doradus in the LMC. In our FTS observations, important FIR cooling lines in the ISM, including CO J=4-3 to 13-12, [CI] 370 micron, and [NII] 205 micron, were clearly detected. In combination with ground-based CO data, we then constructed CO spectral line energy distributions (SLEDs) on 10 pc scales over a 60 pc x 60 pc area and found that the shape of the observed CO SLEDs considerably changes across 30 Doradus, e.g., the peak transition varies from J=6-5 to 10-9, while the slope characterized by the high-to-intermediate J ratio ranges from 0.4 to 1.8. To examine the source(s) of these variations in CO transitions, we analyzed the CO observations, along with [CII] 158 micron, [CI] 370 micron, [OI] 145 micron, H2 0-0 S(3), and FIR luminosity data, using state-of-the-art models of PDRs and shocks. Our detailed modeling showed that the observed CO emission likely originates from highly-compressed (thermal pressure ~ 1e7-1e9 K cm-3) clumps on 0.7-2 pc scales, which could be produced by either UV photons (UV radiation field ~ 1e3-1e5 Mathis fields) or low-velocity C-type shocks (pre-shock medium density ~ 1e4-1e6 cm-3 and shock velocity ~ 5-10 km s-1). Considering the stellar content in 30 Doradus, however, we tentatively excluded the stellar origin of CO excitation and concluded that low-velocity shocks driven by kpc scale processes (e.g., interaction between the Milky Way and the Magellanic Clouds) are likely the dominant source of heating for CO. The shocked CO-bright medium was then found to be warm (temperature ~ 100-500 K) and surrounded by a UV-regulated low pressure component (a few (1e4-1e5) K cm-3) that is bright in [CII] 158 micron, [CI] 370 micron, [OI] 145 micron, and FIR dust continuum emission.

astro-ph.GA↗

arXiv2018

A look at possible microwave dust emission via AKARI infrared all-sky surveys

The anomalous microwave emission (AME) still lacks a conclusive explanation. This excess of emission, roughly between 10 and 50 GHz, correlates spatially with interstellar dust, prompting a "spinning dust" hypothesis: electric dipole emission by rapidly rotating, small dust grains. The typical peak frequency range of the AME profile implicates grains on the order of ~1 nm, suggesting polycyclic aromatic hydrocarbon molecules (PAHs). We compare AKARI/Infrared Camera (IRC), with its thorough PAH-band coverage, to AME intensity estimates from the Planck Collaboration, in the λ Orionis region. We look also at infrared dust emission from other mid IR and far-IR bands. The results and discussion contained here apply to an angular scale of approximately 1°. In general, our results support an AME-from-dust hypothesis. In λ Orionis, we find that certainly dust mass correlates with AME, and that PAH-related emission in the AKARI/IRC 9 μm band may correlate slightly more strongly.

astro-ph.GA↗

arXiv2018

Constraining physical conditions for the PDR of Trumpler 14 in the Carina Nebula

We investigate the physical conditions of the CO gas near the young star cluster, Trumpler 14 of the Carina Nebula. The observations presented in this work are taken with the Fourier Transform Spectrometer (FTS) of the Spectral and Photometric Imaging REceiver (SPIRE) onboard the Herschel Space Observatory. Our field of view covers the edge of a cavity carved by Trumpler 14 about $1\,\mathrm{Myr}$ ago and marks the transition from HII regions to photo-dissociation regions. With the state-of-the-art Meudon PDR code, we successfully derive the physical conditions, which include the thermal pressure ($P$) and the scaling factor of radiation fields ($G_{\mathrm{UV}}$), from the observed CO spectral line energy distributions~(SLEDs) in the observed region. The derived $G_{\mathrm{UV}}$ values generally show an excellent agreement with the UV radiation fields created by nearby OB-stars and thus confirm that the main excitation source of the observed CO emission are the UV-photons provided by the massive stars. The derived thermal pressure is between $0.5-3\,\times\,10^{8}\,\mathrm{K\,cm^{-3}}$ with the highest values found along the ionization front in Car I-E region facing Trumpler 14, hinting that the cloud structure is similar to the recent observations of the Orion Bar. Comparing the derived thermal pressure with the radiation fields, we report the first observationally-derived and spatially-resolved $P \sim 2\times10^4\,G_{\mathrm{UV}}$ relationship. As direct comparisons of the modeling results to the observed $^{13}\mathrm{CO}$, [OI] $63\,\mathrm{μm}$, and [CII] $158\,\mathrm{μm}$ intensities are not straightforward, we urge the readers to be cautious when constraining the physical conditions of PDRs with combinations of $^{12}\mathrm{CO}$, $^{13}\mathrm{CO}$, [CI], [OI] $63\,\mathrm{μm}$, and [CII] $158\,\mathrm{μm}$ observations.

astro-ph.GA↗

arXiv2018

Evolution of the Unidentified Infrared Bands in the Nucleus of the Starburst Galaxy NGC 1097

We present the analysis of the Unidentified Infrared Bands (UIB) in the starburst galaxy NGC 1097. We have combined spectral maps observed with the AKARI/IRC and Spitzer/IRS instruments, in order to study all of the most prominent UIBs, from 3 to 20 micron. Such a complete spectral coverage is crucial to remove the common degeneracies between the effects of the variations of the size distribution and of the charge state of the grains. By studying several UIB ratios, we show evidence that the average size of the UIB carriers is larger in the central region than in the circumnuclear ring. We interpret this result as the selective destruction of the smallest grains by the hard radiation from the central active galactic nucleus.

astro-ph.GA↗

arXiv2017

Near- to mid-Infrared Observations of Galaxy Mergers: NGC2782 and NGC7727

We present the results of near- to mid-infrared (NIR to MIR) imaging and NIR spectroscopic observations of two galaxy mergers, NGC 2782 (Arp 215) and NGC 7727 (Arp 222), with the Infrared Camera on board AKARI. NGC 2782 shows extended MIR emission in the eastern side of the galaxy, which corresponds to the eastern tidal tail seen in the HI 21 cm map, while NGC 7727 shows extended MIR emission in the north of the galaxy, which is similar to the plumes seen in the residual image at the K-band after subtracting a galaxy model. Both extended structures are thought to have formed associated with their merger events. They show excess emission at 7--15 micron, which can be attributed to emission from polycyclic aromatic hydrocarbons (PAHs), while the observed spectral energy distributions decline longward of 24 micron, suggesting that very small grains (VSGs) are deficient. These characteristics of the observed MIR spectral energy distribution may be explained if PAHs are formed by fragmentation of VSGs during merger events. The star formation rate is estimated from the MIR PAH emission in the eastern tail region of NGC 2782 and it is in fair agreement with those estimated from Halpha and [CII] 158 micron. MIR observations are efficient for the study of dust processing and structures formed during merger events.

astro-ph.GA↗

arXiv2016

Disentangling the ISM phases of the dwarf galaxy NGC 4214 using [CII] SOFIA/GREAT observations

The [CII] 158 um fine structure line is one of the dominant cooling lines in the interstellar medium (ISM) and is an important tracer of star formation. Recent velocity-resolved studies with Herschel/HIFI and SOFIA/GREAT showed that the [CII] line can constrain the properties of the ISM phases in star-forming regions. The [CII] line as a tracer of star formation is particularly important in low-metallicity environments where CO emission is weak because of the presence of large amounts of CO-dark gas. The nearby irregular dwarf galaxy NGC 4214 offers an excellent opportunity to study an actively star-forming ISM at low metallicity. We analyzed the spectrally resolved [CII] line profiles in three distinct regions at different evolutionary stages of NGC 4214 with respect to ancillary HI and CO data in order to study the origin of the [CII] line. We used SOFIA/GREAT [CII] 158 um observations, HI data from THINGS, and CO(2-1) data from HERACLES to decompose the spectrally resolved [CII] line profiles into components associated with neutral atomic and molecular gas. We use this decomposition to infer gas masses traced by [CII] under different ISM conditions. Averaged over all regions, we associate about 46% of the [CII] emission with the HI emission. However, we can assign only around 9% of the total [CII] emission to the cold neutral medium (CNM). We found that about 79% of the total molecular hydrogen mass is not traced by CO emission. On average, the fraction of CO-dark gas dominates the molecular gas mass budget. The fraction seems to depend on the evolutionary stage of the regions: it is highest in the region covering a super star cluster in NGC 4214, while it is lower in a more compact, more metal-rich region.

astro-ph.GA↗

arXiv2016

Radiative and mechanical feedback into the molecular gas in the Large Magellanic Cloud. I. N159W

We present Herschel SPIRE Fourier Transform Spectrometer (FTS) observations of N159W, an active star-forming region in the Large Magellanic Cloud (LMC). In our observations, a number of far-infrared cooling lines including CO(4-3) to CO(12-11), [CI] 609 and 370 micron, and [NII] 205 micron are clearly detected. With an aim of investigating the physical conditions and excitation processes of molecular gas, we first construct CO spectral line energy distributions (SLEDs) on 10 pc scales by combining the FTS CO transitions with ground-based low-J CO data and analyze the observed CO SLEDs using non-LTE radiative transfer models. We find that the CO-traced molecular gas in N159W is warm (kinetic temperature of 153-754 K) and moderately dense (H2 number density of (1.1-4.5)e3 cm-3). To assess the impact of the energetic processes in the interstellar medium on the physical conditions of the CO-emitting gas, we then compare the observed CO line intensities with the models of photodissociation regions (PDRs) and shocks. We first constrain the properties of PDRs by modelling Herschel observations of [OI] 145, [CII] 158, and [CI] 370 micron fine-structure lines and find that the constrained PDR components emit very weak CO emission. X-rays and cosmic-rays are also found to provide a negligible contribution to the CO emission, essentially ruling out ionizing sources (ultraviolet photons, X-rays, and cosmic-rays) as the dominant heating source for CO in N159W. On the other hand, mechanical heating by low-velocity C-type shocks with ~10 km/s appears sufficient enough to reproduce the observed warm CO.

astro-ph.GA↗