In this paper we describe the requirements for research information systems
and problems which arise in the development of such system. Here is shown which
problems could be solved by using of knowledge markup technologies. Ontology
for Research Information System offered. Architecture for collecting research
data and providing access to it is described.
Authors' comments: 8 pages, ontology description included, position paper at the
Workshop on Knowledge Markup and Semantic Annotation at K-CAP'2001
This decade has seen a great deal of progress in the development of
information retrieval systems. Unfortunately, we still lack a systematic
understanding of the behavior of the systems and their relationship with
documents. In this paper we present a completely new approach towards the
understanding of the information retrieval systems. Recently, it has been
observed that retrieval systems in TREC 6 show some remarkable patterns in
retrieving relevant documents. Based on the TREC 6 observations, we introduce a
geometric linear model of information retrieval systems. We then apply the
model to predict the number of relevant documents by the retrieval systems. The
model is also scalable to a much larger data set. Although the model is
developed based on the TREC 6 routing test data, I believe it can be readily
applicable to other information retrieval systems. In Appendix, we explained a
simple and efficient way of making a better system from the existing systems.
Authors' comments: 13 pages
This paper proposes a Japanese/English cross-language information retrieval
(CLIR) system targeting technical documents. Our system first translates a
given query containing technical terms into the target language, and then
retrieves documents relevant to the translated query. The translation of
technical terms is still problematic in that technical terms are often compound
words, and thus new terms can be progressively created simply by combining
existing base words. In addition, Japanese often represents loanwords based on
its phonogram. Consequently, existing dictionaries find it difficult to achieve
sufficient coverage. To counter the first problem, we use a compound word
translation method, which uses a bilingual dictionary for base words and
collocational statistics to resolve translation ambiguity. For the second
problem, we propose a transliteration method, which identifies phonetic
equivalents in the target language. We also show the effectiveness of our
system using a test collection for CLIR.
Authors' comments: 9 pages, 5 Postscript figures, uses colacl.sty and psfig.tex
Texts exhibit considerable stylistic variation. This paper reports an
experiment where a corpus of documents (N= 75 000) is analyzed using various
simple stylistic metrics. A subset (n = 1000) of the corpus has been previously
assessed to be relevant for answering given information retrieval queries. The
experiment shows that this subset differs significantly from the rest of the
corpus in terms of the stylistic metrics studied.
Authors' comments: Proceedings of NEMLAP-2
We study the model of massless $1+1$ electrodynamics with nonconstant
coupling, introduced by Peet, Susskind and Thorlacius as the `charge hole'. But
we take the boundary of the strong coupling region to be first timelike, then
spacelike for a distance $X$, and then timelike again (to mimic the structure
of a black hole). For an incident charge pulse entering this `charge trap' the
charge and information get separated. The charge comes out near the endpoint of
the singularity. The `information' travels a well localised path through the
strong coupling region and comes out later.
Authors' comments: 21 pages, 1 figure, harvmac
With the increasing demand for step-wise, cross-modal, and knowledge-grounded reasoning, multimodal large language models (MLLMs) are evolving beyond the traditional fixed retrieve-then-generate paradigm toward more sophisticated agentic multimodal retrieval-augmented generation (MM-RAG). Existing benchmarks, however, mainly focus on simplified QA with short retrieval chains, leaving adaptive planning and multimodal reasoning underexplored. We present MC-Search, the first benchmark for agentic MM-RAG with long, step-wise annotated reasoning chains spanning five representative reasoning structures. Each example specifies sub-questions, retrieval modalities, supporting facts, and intermediate answers, with fidelity ensured by HAVE (Hop-wise Attribution and Verification of Evidence), resulting in 3,333 high-quality examples averaging 3.7 hops. Beyond answer accuracy, MC-Search introduces new process-level metrics for reasoning quality, stepwise retrieval and planning accuracy. By developing a unified agentic MM-RAG pipeline, we benchmark six leading MLLMs and reveal systematic issues such as over- and under-retrieval and modality-misaligned planning. Finally, we introduce Search-Align, a process-supervised fine-tuning framework leveraging verified reasoning chains, showing that our data not only enables faithful evaluation but also improves planning and retrieval fidelity in open-source MLLMs.
Authors' comments: ICLR 2026
4267.
On elliptic systems with $k$-wise interactions in the strong competition regime: uniform Hölder bounds and properties of the limiting configurations
In this paper we investigate a class of variational reaction-diffusion systems with strong competition driven by beyond-pairwise interactions. The model involves $d$ nonnegative components interacting through $k$-wise terms, with $3 \leq k \leq d$, and includes symmetric interaction coefficients accounting for multi-component effects as well as suitable nonlinear terms. We focus on minimal energy solutions, proving uniform-in-$β$ Hölder bounds up to an explicit threshold exponent depending only on the dimension of the space and on the order $k$ of the interaction. As $β\to +\infty$, we show that minimizers converge strongly in $H^1$ and in Hölder spaces to a partially segregated configuration, characterized as minimizer of a natural variational problem under a $k$-segregation constraint. Finally, we prove that every minimizer of the limit problem enjoys the Hölder regularity and we derive some basic extremality conditions.
Authors' comments: 35 pages. arXiv admin note: text overlap with arXiv:2409.11976 by other authors
4268.
LAF-YOLOv10 with Partial Convolution Backbone, Attention-Guided Feature Pyramid, Auxiliary P2 Head, and Wise-IoU Loss for Small Object Detection in Drone Aerial Imagery
Sohail Ali Farooqui, Zuhair Ahmed Khan Taha, Mohammed Mudassir Uddin, Shahnawaz Alam
Unmanned aerial vehicles serve as primary sensing platforms for surveillance, traffic monitoring, and disaster response, making aerial object detection a central problem in applied computer vision. Current detectors struggle with UAV-specific challenges: targets spanning only a few pixels, cluttered backgrounds, heavy occlusion, and strict onboard computational budgets. This study introduces LAF-YOLOv10, built on YOLOv10n, integrating four complementary techniques to improve small-object detection in drone imagery. A Partial Convolution C2f (PC-C2f) module restricts spatial convolution to one quarter of backbone channels, reducing redundant computation while preserving discriminative capacity. An Attention-Guided Feature Pyramid Network (AG-FPN) inserts Squeeze-and-Excitation channel gates before multi-scale fusion and replaces nearest-neighbor upsampling with DySample for content-aware interpolation. An auxiliary P2 detection head at 160$\times$160 resolution extends localization to objects below 8$\times$8 pixels, while the P5 head is removed to redistribute parameters. Wise-IoU v3 replaces CIoU for bounding box regression, attenuating gradients from noisy annotations in crowded aerial scenes. The four modules address non-overlapping bottlenecks: PC-C2f compresses backbone computation, AG-FPN refines cross-scale fusion, the P2 head recovers spatial resolution, and Wise-IoU stabilizes regression under label noise. No individual component is novel; the contribution is the joint integration within a single YOLOv10 framework. Across three training runs (seeds 42, 123, 256), LAF-YOLOv10 achieves 35.1$\pm$0.3\% mAP@0.5 on VisDrone-DET2019 with 2.3\,M parameters, exceeding YOLOv10n by 3.3 points. Cross-dataset evaluation on UAVDT yields 35.8$\pm$0.4\% mAP@0.5. Benchmarks on NVIDIA Jetson Orin Nano confirm 24.3 FPS at FP16, demonstrating viability for embedded UAV deployment.
4269.
27 years of Spaceborne IR Astronomy: An ISO, Spitzer, WISE and NEOWISE Survey for Large-Amplitude Variability in Young Stellar Objects
Chinmay S. Kulkarni, Thomas Behling, Elisabeth E. Banks, Jason Jones, Tyler Robbins, Nathanael Burns-Watson, S. Thomas Megeath, Robert Gutermuth et al.
Infrared observations can probe photometric variability across the full evolutionary range of young stellar objects (YSOs), from deeply embedded protostars to pre-main-sequence stars with dusty disks. We present 3-8 micron light curves extending 27 years from 1997 to 2024 obtained with three space-based IR telescopes: ISO, Spitzer and WISE. Although unevenly sampled with large gaps in coverage, these light curves show variability on time scales ranging from days to decades. We focus on the Spitzer-identified YSOs with disks and envelopes that exhibit variations of a factor of two or more in this wavelength range. We identified seven YSOs where the light curves are dominated by bursts of sustained (> 5 yr) high flux, including four that show a steep decay ending the burst and three that are ongoing as of the final observation. We find six YSOs that are undergoing declines, which may be the end of bursts that began before 1997. The most common form of variability, exhibited by 26 YSOs in our sample, show variations over time intervals of years to months but do not exhibit sustained bursts or fades. The Spitzer [3.6]-[4.5] and WISE [3.5]-[4.6] colors either increase or remain constant with increasing brightness, inconsistent with dust extinction as being the primary source of the large-amplitude variability.
Authors' comments: 69 pages, 11 main figures, 3 figures and 5 figure sets in appendix. Accepted for publication in The Astrophysical Journal (ApJ)
4270.
Revisiting "Revisiting Neuron Coverage for DNN Testing: A Layer-Wise and Distribution-Aware Criterion": A Critical Review and Implications on DNN Coverage Testing
Jinhan Kim, Nargiz Humbatova, Gunel Jahangirova, Shin Yoo, Paolo Tonella
We present a critical review of Neural Coverage (NLC), a state-of-the-art DNN coverage criterion by Yuan et al. at ICSE 2023. While NLC proposes to satisfy eight design requirements and demonstrates strong empirical performance, we question some of their theoretical and empirical assumptions. We observe that NLC deviates from core principles of coverage criteria, such as monotonicity and test suite order independence, and could more fully account for key properties of the covariance matrix. Additionally, we note threats to the validity of the empirical study, related to the ground truth ordering of test suites. Through our empirical validation, we substantiate our claims and propose improvements for future DNN coverage metrics. Finally, we conclude by discussing the implications of these insights.
Authors' comments: ICSE 2026
4271.
The JWST weather report from the nearest brown dwarfs III: Heterogeneous clouds and Thermochemical instabilities as possible drivers of WISE 1049AB's spectroscopic variability
Natalia Oliveros-Gomez, Elena Manjavacas, Theodora Karalidi, Myrla Phillippe, Beatriz Campos Estrada, Beth Biller, Johanna M. Vos, Jacqueline Faherty et al.
We present a new analysis of the spectroscopic variability of WISE~J104915.57$-$531906.1AB (WISE~1049AB, L7.5+T0.5), observed using the NIRSpec instrument onboard the James Webb Space Telescope (GO 2965 - PI: Biller). We explored the variability of the dominant molecular bands present in their 0.6--5.3~$μ$m spectra (H$_2$O, CH$_4$, CO), finding that the B component exhibits a higher maximum deviation than the A component in all the wavelength ranges tested. The light curves reveal wavelength-(atmospheric depth) and possibly chemistry-dependent variability. In particular, for the A component, the variability in the light curves at the wavelengths traced by the CH$_4$ and CO molecular absorption features is higher than that of H$_2$O, even when both trace similar pressure levels. We concluded that clouds alone are unlikely to explain the increased variability of CO and CH$_4$ with respect to H$_2$O, suggesting that an additional physical mechanism is needed to explain the observed variability. This mechanism is probably due to thermochemical instabilities. Finally, we provide a visual representation of the 3D atmospheric map reconstructed for both components using the molecular band contributions at different pressure levels and the fit of planetary-scale waves.
Authors' comments: Accepted for publication in the Astrophysical Journal (ApJ). 25 pages, 11 figures, 3 tables
4272.
Reconstruction of the non-linear wave at a buoy from shoreline data and
applications to the tsunami inverse problem for piece-wise sloping bathymetry
We discuss the following inverse problem: given the run-up data of a tsunami
wave, can we recover its initial shape? We study this problem within the
framework of the non-linear shallow water equations, a model widely used to
study tsunami propagation and inundation. Previously, it has been demonstrated
that in the case of infinite sloping bathymetry, it is possible to recover the
initial water displacement and velocity from shoreline readings
\cite{Rybkin23,Rybkin24,Rybkin25}.
We consider a finite sloping bathymerty. We show that it is possible to
recover boundary conditions (water displacement and velocity) on a virtual buoy
from the shoreline data. Further, we discuss stitching together the shallow
water equations and the Boussinesq equation in a more complex piece-wise
sloping bathymetry in order to recover the initial conditions, while
incorporating the dispersion to our model.
4273.
Hot Diggity Dog: Simultaneous CO and Dust Modeling of the Most Luminous WISE Hot Dust-Obscured Galaxy Unveils Extreme Molecular Gas Excitation
Kevin C. Harrington, Román Fernández Aranda, Leindert Boogaard, Axel WeiÃ, Tanio Diaz Santos, Manuel Aravena, Roberto J. Assef, Chao-Wei Tsai et al.
Hot dust-obscured galaxies (Hot DOGs), the most infrared (IR) luminous objects selected by the WISE all-sky mid-IR survey, have yielded a sample of intrinsically luminous quasars (QSOs) with obscured nuclear activity and hot dust temperatures. The molecular gas excitation properties have yet to be examined in detail under such extreme conditions. Here we study the most far-IR luminous \textit{WISE} Hot DOG W2246-0526, focusing on ALMA observations of the central host galaxy. Multi-J CO transition measurements at J=2-1, 5-4, 7-6, 12-11, and 17-16 provide the first self-consistent modeling constraints on the molecular gas and dust properties of any WISE Hot DOG to date, providing a benchmark for future studies of dust-obscured QSOs. We implement a state-of-the-art TUrbulent Non-Equilibrium Radiative transfer model (TUNER) that simultaneously models both the line and dust continuum measurements. The extreme CO spectral line energy distribution seems to turnover around the CO(12-11) transition, likely making this among the most highly excited galaxies ever reported. The model infers a molecular gas mass $\sim 8\times10^{10}$\,\Msun, and we conclude that J$\ge$3 CO line luminosities trace the bulk of the molecular gas mass for this extreme system unlike low-excitation conditions calibrated locally. W2246-0526 is a rapidly evolving system, with a high mean value of the molecular gas kinetic temperature versus dust temperature $T_\mathrm{k}$ / $T_\mathrm{d} \sim 4.3$. This may be due to the shocks and outflows injecting kinetic energy within the central kpc of this host that have previously been reported. These signs of highly excited molecular gas in W2246-0526 motivates obtaining well-sampled CO ladders in larger samples to better understand the conditions within these short-lived episodes associated with the most obscured supermassive black hole activity.
Authors' comments: Accepted by A&A. 7 pages, 4 Appendix pages, 6 figures
Vapor condensation is a physical phenomenon that finds application in heat
removal systems. The traditional design of these systems involves round tubes
but experience shows that this geometry is not optimal for heat transfer.
Flattened tubes on the other hand, have been found to offer potential for
improvement as their geometry increases the condensation surface, which fosters
higher heat transfer rates. However, the effects of tube shape (aspect ratio)
and orientation (rotation angle) on film-wise condensation dynamics are not
fully understood. In this work, we numerically simulate a model of the
condensed vapor layer thickness distribution on the flattened tube inner
surfaces taking into account bulk and surface forces (gravity, surface tension,
shear stress) for a thin layer of liquid. We consider various configurations of
aspect ratios (circular, and AR = 2, 4, and 6) and rotation angles (0{\deg},
10{\deg}, 20{\deg}, 30{\deg}, 45{\deg}, 60{\deg}, 75{\deg}, and 90{\deg}). Our
simulations allow for an improved understanding of how these geometric
parameters as well as their interplay, influence the thickness distribution of
the condensate film on the tube's inner surface, and facilitate the
identification of configurations that maximize heat transfer efficiency.
Considering water as a working fluid, results show a possible heat transfer
enhancement of up to 74% compared to the round tube geometry for an aspect
ratio of 6 and a rotation angle of 90{\deg}.
4275.
A Novel Compression Framework for YOLOv8: Achiev-ing Real-Time Aerial
Object Detection on Edge Devices via Structured Pruning and Channel-Wise
Distillation
Melika Sabaghian, Mohammad Ali Keyvanrad, Seyyedeh Mahila Moghadami
Efficient deployment of deep learning models for aerial object detection on
resource-constrained devices requires significant compression without
com-promising performance. In this study, we propose a novel three-stage
compression pipeline for the YOLOv8 object detection model, integrating
sparsity-aware training, structured channel pruning, and Channel-Wise Knowledge
Distillation (CWD). First, sparsity-aware training introduces dynamic sparsity
during model optimization, effectively balancing parameter reduction and
detection accuracy. Second, we apply structured channel pruning by leveraging
batch normalization scaling factors to eliminate redundant channels,
significantly reducing model size and computational complexity. Finally, to
mitigate the accuracy drop caused by pruning, we employ CWD to transfer
knowledge from the original model, using an adjustable temperature and loss
weighting scheme tailored for small and medium object detection. Extensive
experiments on the VisDrone dataset demonstrate the effectiveness of our
approach across multiple YOLOv8 variants. For YOLOv8m, our method reduces model
parameters from 25.85M to 6.85M (a 73.51% reduction), FLOPs from 49.6G to
13.3G, and MACs from 101G to 34.5G, while reducing AP50 by only 2.7%. The
resulting compressed model achieves 47.9 AP50 and boosts inference speed from
26 FPS (YOLOv8m baseline) to 45 FPS, enabling real-time deployment on edge
devices. We further apply TensorRT as a lightweight optimization step. While
this introduces a minor drop in AP50 (from 47.9 to 47.6), it significantly
improves inference speed from 45 to 68 FPS, demonstrating the practicality of
our approach for high-throughput, re-source-constrained scenarios.
Authors' comments: 28 pages, 11 figures
4276.
A Voxel-Wise Uncertainty-Guided Framework for Glioma Segmentation Using
Spherical Projection-Based U-Net and Localized Refinement in Multi-Parametric
MRI
Purpose: Accurate segmentation of glioma subregions in multi-parametric MRI
(MP-MRI) is essential for diagnosis and treatment planning but remains
challenging due to tumor heterogeneity and ambiguous boundaries. This study
proposes an uncertainty-guided hybrid framework integrating spherical
projection-based 2D modeling with targeted 3D refinement to enhance
segmentation accuracy and interpretability. Methods: Using the BraTS2020
dataset (369 patients, four-modality MP-MRI), three 2D U-Nets were trained to
segment enhancing tumor (ET), tumor core (TC), and whole tumor (WT). Voxel-wise
uncertainty was quantified via a spherical projection-based 2D nnU-Net,
capturing prediction variance across deformed inputs. A 3D sliding window was
used to identify high-uncertainty regions, which were refined using a dedicated
3D nnU-Net. Final outputs combined 2D and 3D predictions through a weighted
fusion optimized via Particle Swarm Optimization. Results: The proposed method
outperformed standalone 2D and 3D baselines, achieving Dice scores of 0.8124
(ET), 0.7499 (TC), and 0.9055 (WT), with consistent gains in sensitivity and
visual coherence. Conclusion: This work presents a novel uncertainty-aware
segmentation strategy that adaptively integrates 2D and 3D modeling. By
focusing refinement on ambiguous regions, it improves both efficiency and
accuracy, offering broad applicability to precision neuro-oncology and other
high-stakes medical imaging tasks.
Authors' comments: 31 pages, 6 figures, 2 tables
4277.
Alignment Quality Index (AQI) : Beyond Refusals: AQI as an Intrinsic
Alignment Diagnostic via Latent Geometry, Cluster Divergence, and Layer wise
Pooled Representations
Alignment is no longer a luxury, it is a necessity. As large language models
(LLMs) enter high-stakes domains like education, healthcare, governance, and
law, their behavior must reliably reflect human-aligned values and safety
constraints. Yet current evaluations rely heavily on behavioral proxies such as
refusal rates, G-Eval scores, and toxicity classifiers, all of which have
critical blind spots. Aligned models are often vulnerable to jailbreaking,
stochasticity of generation, and alignment faking.
To address this issue, we introduce the Alignment Quality Index (AQI). This
novel geometric and prompt-invariant metric empirically assesses LLM alignment
by analyzing the separation of safe and unsafe activations in latent space. By
combining measures such as the Davies-Bouldin Score (DBS), Dunn Index (DI),
Xie-Beni Index (XBI), and Calinski-Harabasz Index (CHI) across various
formulations, AQI captures clustering quality to detect hidden misalignments
and jailbreak risks, even when outputs appear compliant. AQI also serves as an
early warning signal for alignment faking, offering a robust, decoding
invariant tool for behavior agnostic safety auditing.
Additionally, we propose the LITMUS dataset to facilitate robust evaluation
under these challenging conditions. Empirical tests on LITMUS across different
models trained under DPO, GRPO, and RLHF conditions demonstrate AQI's
correlation with external judges and ability to reveal vulnerabilities missed
by refusal metrics. We make our implementation publicly available to foster
future research in this area.
4278.
Why is distortion inevitable in opinion propagation on social media?
Noise induced layer-wised synchronization in Noise-Frustrated
Hegselmann-Krause model
The proliferation of social media as a dominant information propagation
platform has intensified scholarly concerns about systemic information
distortion,a phenomenon where content undergoes progressive alteration during
multi layered transmission. However, existing literature extensively documents
distortion patterns, the fundamental mechanisms coupling network architecture
with cognitive noise remain poorly quantified. Here, we introduce a novel
fractal network with coupled Noise Frustrated Hegselmann Krause (NFHK)
framework that systematically disentangles these intertwined factors. By
integrating fractal topology analysis with modified bounded confidence
dynamics, our model reveals how hierarchical network structures (characterized
by scale invariant connectivity patterns) amplify stochastic noise through
successive retransmission layers. Through rigorous mathematical analysis, multi
agent simulations, and empirical validation of typical retweet cascades, we
demonstrate two key phenomena: (i) distortion escalates super linearly with
network depth and (ii) peer nodes exhibit emergent layer wise synchronization
despite lacking direct connections among themselves and form a number of
synchronous groups based on the number of network layers. These findings
establish a unified mechanism explaining distortion accumulation in digital
ecosystems while challenging conventional "echo chamber" narratives. Our noise
frustration protocol can offer actionable insights for policymakers to design
topology aware regulatory frameworks. This work bridges complex systems theory
with computational social science, providing both a mathematical foundation for
distortion analysis and a toolkit for platform governance.
4279.
The JWST weather report from the nearest brown dwarfs II: Consistent
variability mechanisms over 7 months revealed by 1-14 $μ$m NIRSpec + MIRI
monitoring of WISE 1049AB
Xueqing Chen, Beth A. Biller, Xianyu Tan, Johanna M. Vos, Yifan Zhou, Genaro Suárez, Allison M. McCarthy, Caroline V. Morley et al.
We present a new epoch of JWST spectroscopic variability monitoring of the
benchmark binary brown dwarf WISE 1049AB, the closest, brightest brown dwarfs
known. Our 8-hour MIRI low resolution spectroscopy (LRS) and 7-hour NIRSpec
prism observations extended variability measurements for any brown dwarfs
beyond 11 $\mu$m for the first time, reaching up to 14 $\mu$m. Combined with
the previous epoch in 2023, they set the longest JWST weather monitoring
baseline to date. We found that both WISE 1049AB show wavelength-dependent
light curve behaviours. Using a robust k-means clustering algorithm, we
identified several clusters of variability behaviours associated with three
distinct pressure levels. By comparing to a general circulation model (GCM), we
identified the possible mechanisms that drive the variability at these pressure
levels: Patchy clouds rotating in and out of view likely shaped the dramatic
light curves in the deepest layers between 1-2.5 $\mu$m, whereas hot spots
arising from temperature / chemical variations of molecular species likely
dominate the high-altitude levels between 2.5-3.6 $\mu$m and 4.3-8.5 $\mu$m.
Small-grain silicates potentially contributed to the variability of WISE 1049A
at 8.5-11 $\mu$m. While distinct atmospheric layers are governed by different
mechanisms, we confirmed for the first time that each variability mechanism
remains consistent within its layer over the long term. Future multi-period
observations will further test the stability of variability mechanisms on this
binary, and expanded JWST variability surveys across the L-T-Y sequence will
allow us to trace and understand variability mechanisms across a wider
population of brown dwarfs and planetary-mass objects.
Authors' comments: 21 pages, 18 figures, accepted for publication in MNRAS
Large language models (LLMs) hold great promise for specialized scientific
domains such as materials science, yet adapting them efficiently and accurately
to domain-specific knowledge remains challenging due to limited data and high
knowledge density. We propose a two-stage framework that combines structured
model compression with a scientific fine-tuning regimen to address this
challenge. In the compression stage, we decompose the LLM's weight matrices
into local low-rank "rank blocks" and arrange these blocks in a Penrose-like
non-periodic tiling pattern. Each block is then compacted via spectral
transformations (e.g., discrete cosine or Fourier transforms), and a
Kullback-Leibler (KL) divergence-based alignment loss preserves the
distributional similarity between the compressed model's representations and
those of the original full model. In the adaptation stage, the compressed model
is further tuned using a human-like scientific reading protocol: it processes
technical materials science documents section by section, engaging in a
structured question-and-answer routine for each section. This section-wise Q&A
fine-tuning strategy extracts explicit reasoning traces and gradually injects
domain knowledge, while minimizing catastrophic forgetting of the model's
general language capabilities. By balancing efficient compression with targeted
adaptation, our two-stage approach enables precise specialization of LLMs to
high-value domains under data-scarce conditions. We present this principled yet
exploratory pipeline and outline its potential for advancing materials science
knowledge integration, laying the groundwork for comprehensive empirical
evaluation in future work.
