메커니즘적 편집(어블레이션, 가중치 편집, 활성화 스티어링)은 유용한 능력을 보존하면서 신경망에서 유해한 능력을 언러닝(unlearning)하기 위한 표준적인 도구이다. 현재의 접근법들은 테스트를 통해서만 그 효과를 검증하는데, 이는 입력의 전체 연속 영역을 결코 다 다룰 수 없다. 해석가능성(interpretability)과 검증(verification)의 경계에 있는 선행 연구들은 모델에 대한 설명을 인증한다: 회로가 무엇을 계산하는지, 또는 그것이 전체를 충실하게 설명하는지 여부이다. 우리는 대신 편집의 행동적 효과를 인증한다: 즉 회로를 비활성화하는 것이 한 스킬을 제거하면서도 다른 스킬을 해당 영역의 모든 입력에 대해 입증 가능하게 보존한다는 것이며, 이는 정보흐름보안(information-flow security) 의미에서의 특징 비간섭(feature non-interference) 보장에 해당한다.
우리는 토이(toy) 렐루 네트워크부터 표준적인 소프트맥스+레이어놈(LayerNorm) 트랜스포머에 이르기까지 이러한 인증된 편집을 시연하며, 연속적인 임베딩 공간 영역에 대해 제거와 보존을 증명하고, 안전한(sound) 바운드 전파(bound propagation)로 전환함으로써 정확한(exact) 솔버가 다룰 수 있는 입력 교란 차원의 약 9배에 도달한다. 더 나아가 우리는 어떠한 유한한 결정론적 블랙박스 테스트도 제거를 인증할 수 없음을 증명하며, 전수 테스트를 통과하지만 임의로 작게 만들 수 있는 생존 영역(survivor pocket)에서 입증 가능하게 실패하는 편집 사례를 제시한다. 이러한 보장은 작은 표준 아키텍처 네트워크에서 유효하며, 다른 모든 제거 주장과 마찬가지로 대상 스킬이 결정 가능한 명세(decidable specification)를 허용한다는 것을 전제로 하는데, 이는 실제 세계의 유해성이 갖지 못할 수도 있는 속성이다.
Current analyses of LLMs' parametric knowledge are largely output-centric, drawing conclusions about what a model knows without verifying what it was actually trained on. This leaves fundamental questions, such as whether a correct response reflects genuine generalization or rote memorization, grounded in speculation rather than evidence. To resolve these ambiguities, we introduce LUMOS, a diagnostic framework that traces knowledge across training-data exposure, internal representations, and behavioral outputs, leveraging OLMo 2 with its fully transparent training corpus. By grounding analysis in verified exposure, we reveal that models internally encode rare facts with high separability (84%) yet fail to express them behaviorally (54%), though this retrieval gap narrows with scale. Furthermore, when models are asked to self-reflect on their own answers, they perform reliably on trained content (83%) but drop to random-baseline levels (49%) on unseen content. This collapse persists even under chain-of-thought prompting, which inflates confidence signals rather than improving calibration. Collectively, these findings demonstrate that incorporating the training-data axis into LLM evaluation transforms speculative diagnoses into verifiable claims, and we advocate that this axis should be a standard component of knowledge assessment in LLMs.
Low-rank adaptation (LoRA) provides a parameter-efficient approach to continual learning, but its nominal rank can conceal a loss of effective adaptation capacity. We identify \emph{spectral plasticity collapse}: during sequential adaptation, update energy becomes concentrated in a small subset of singular modes, leaving much of the available low-rank space underutilized. This exposes a limitation of interference avoidance alone: protecting historical representations does not ensure that the remaining adaptation capacity is responsive to new tasks or effectively utilized. To address this problem, we propose \texttt{MLoRA}, which jointly controls capacity allocation and utilization. First, historical whitening identifies input directions with strong current-task response relative to accumulated historical response, yielding a task-adaptive basis that remains fixed during training. Second, approximate polar orthogonalization of momentum updates reduces spectral concentration within theselected space. An orthonormal basis connects these mechanisms by transferring the factor-update spectrum exactly tothe induced weight update. We establish a max--min characterization of exact subspace selection and derive cumulative spectral bounds under controlled cross-step anisotropy. Across five class-incremental benchmarks and eight incremental settings, \texttt{MuLoRA} achieves the highest mean accuracy in 15 of 16 reported metrics.
Natural language has become an interface to code, images and molecules, but not to living matter: a stimulus applied to cells has no fixed linguistic meaning, and every paired example of an instruction and its outcome costs a wet-lab experiment. We ask how far an existing archive of experiments can take this interface, using xenobots, motile multicellular constructs made of frog cells, stimulated by a single electrical pulse of varying duration. We re-track 155 archived recordings with two independent methods, measure what the pulse does from each bot's own behaviour before and after it, and separate behaviours the pulse controls from behaviours it does not. Longer pulses stop bots and short pulses leave them moving, sometimes faster; turning direction and starting from rest do not depend on duration, and restarting at most weakly. We then replace hand-made behaviour labels with a blind language critic: fresh sessions of a multimodal language model read plots of each bot and score how well its behaviour matches each of 176 instructions. The critic agrees with the measured labels whenever its plots contain the information the labels are defined on, and dose curves learned from its scores match those learned from the measurements. A prompt-to-intervention model trained only on critic scores maps unseen instructions, including negations, to the right end of the dose range on unseen bots, as well as the same model trained on measured labels. The archive therefore supports a calibrated, refusing language interface for the three behaviours a pulse can set, and identifies, rather than hides, the behaviours it cannot.
How complex can the responses of a quantum process become as it runs longer with fixed internal memory? We measure this complexity by sequential response capacity, the depth of an adaptive tree of experiments whose acceptance probabilities separate possible processes by a prescribed gap. Each experiment uses a fresh complete run, and one fixed process must realize every response along a path. For any fixed common visible dimension of at least two and internal memory cap, we prove that capacity has order $K\log((K+1)/\gamma)$ for $K$ transitions per run and probability gap $0<\gamma\leq1$. Time-dependent single-qubit phase channels without internal memory already attain the fixed-gap $K\log K$ growth, realizing each path with zero-or-one response probabilities. Under the same tester permissions, classical stochastic processes that measure in a fixed basis at every transition have only linear capacity at fixed dimensions and gap. For phase sequences selected by a classical address stored inside the process, we then quantify how known independent Pauli noise changes this logarithmic enhancement. With ideal tester controls and weak residual phase noise after correction, we prove matching capacity bounds at a fixed small gap. At fixed address size and fixed positive residual noise, the logarithmic enhancement saturates once the phase-sequence length exceeds the inverse noise scale, and capacity then grows linearly with sequence length.
Byte-level byte-pair encoding (BBPE) tokenizers are attractive for multilingual large language models because they cover all Unicode text. In UTF-8-based BBPE, however, many scripts start from a higher fallback cost than English: when no learned merges apply, a multibyte character requires multiple byte-derived symbols. We call this worst-case pre-merge cost the encoding floor. A higher floor can increase token counts and per-request cost while shrinking usable context. Changing the whole text encoding can reduce this gap but make already-efficient English spans more expensive in mixed-script text. Thus, we propose Universal Byte-Level Encoding (UBE), a dual-alphabet tokenizer that keeps 1-2-byte UTF-8 characters in UTF-8 and routes 3-4-byte characters through UTF-16. This lowers the encoding floor for 3-byte Basic Multilingual Plane characters, used by many scripts that need more tokens than English, without raising it for any character. UBE changes only the symbols presented to byte-pair encoding, and a Unicode 17 audit confirms exact round trips for every scalar value and every input in three official test suites. UBE also composes with other boundary policies and morphology-based encodings. Across intrinsic evaluations, UBE lowers dispersion in English-normalized token-count ratios (Gini 6-14% lower on FLORES-200 at 16K-64K), reducing cross-lingual token-budget disparity. In multilingual language models, UBE matches BBPE's quality. In the main multilingual settings, UBE reduces token counts most where they are highest and also lowers English counts, yielding more usable context and faster prompt processing in content-matched benchmarks. Byte-level models benefit directly: UBE shortens their inputs by 29-33% in all 39 of the FLORES-200 languages written mostly in 3-byte characters (1.4-1.5 times as much text per context window), lowers byte-level Gini by 28%, and lengthens no input.
누가 또는 무엇이 의식을 가지고 있는가? 주관적 경험은 1인칭 관점에서만 직접 접근할 수 있기 때문에, 다른 개체의 의식에 대한 판단은 부분적으로 유추(analogy)에 의존한다. 역사적으로 이러한 추론은 인간이 아닌 동물에 집중되어 왔으나, 인공지능(AI)의 발전으로 의식을 가진 AI의 가능성이 제기되었다. 본 논문에서는 이러한 증거적 유추를 평가하기 위한 인과적 프레임워크를 개발한다. 핵심적인 구분은 의식 생성에 관여할 가능성이 높은 요인들의 유사성과 하류(downstream) 행동 또는 인지적 효과에서의 유사성 사이의 차이다. 제안하는 프레임워크는 알려지지 않은 원인, 무력화 요인(disabling differences), 의식에 이르는 대안적 경로를 고려하면서 인과적 관련성에 따라 원천-목표(source-target) 유사성에 가중치를 부여한다. 생물학적 시스템에 적용했을 때, 이는 인간과의 인과적 거리가 멀어질수록 유추적 지지가 전반적으로 약화되는 이유를 설명한다. 현대 AI에 적용했을 때, 이는 관련 인과적 대응 관계가 여전히 명확히 확립되지 않았기 때문에 행동적 유사성만으로는 의식에 대한 제한적인 증거만을 제공함을 시사한다. 또한 이 프레임워크는 어떤 증거가 인공의식에 대한 주장을 강화할 수 있는지를 명확히 밝힌다.
Large language models generate candidate programs for automatic heuristic design, while evolutionary search evaluates and selects them to guide further generation. Online post-training uses rewards from candidate evaluations to adapt the generator. Retained programs can also guide later generation as parents. This motivates rewards that reflect their reuse opportunities beyond immediate fitness. We propose Search-value credit (SVC), which constructs response-level rewards from actual admission to the retained set and a short-term parent-exposure proxy. Admission determines reward eligibility; the proxy weights eligible responses by reuse opportunities under existing parent-selection rules. Rewards are computed after candidate evaluation, without evaluating descendants. SVC retains the search and learning procedures of the Co-Evolution of Algorithms and Language Model (CALM) framework. Across four combinatorial optimization tasks with the same base model and response budget, SVC achieves the best mean final search performance on three tasks among six conditions. Under identical prompts and parents on traveling salesman and capacitated vehicle routing tasks, SVC-trained generators yield a higher mean proportion of responses whose programs outperform the best prompting parent than those trained with CALM's native reward. These findings support reward construction using search retention and selection information.
Score-based variational inference (SBVI) provides an alternative to classical Kullback--Leibler (KL)-based variational inference by learning a distribution using only score queries of the target. EigenVI~\cite{cai2024eigenvi} formulates SBVI as an eigenvalue problem, but faces two limitations: the number of variational parameters grows exponentially with the data dimension, and the single-eigenvector formulation cannot represent the full minimum-eigenvalue eigenspace. We propose QuanVI, a quantum-inspired SBVI algorithm that addresses both limitations. Rather than selecting a single eigenvector, QuanVI captures a learned subspace through a density-operator formulation. To enable scalable computation, \model{} employs a matrix-product-operator (MPO) structure parameterized by a quantum tensor network (QTN) with unitary-constrained local tensors and auxiliary ancilla wires. Numerical experiments demonstrate the scalability of QuanVI beyond EigenVI and competitive performance on synthetic distributions and Bayesian posterior-approximation benchmarks, including challenging non-Gaussian targets. The code is available at https://github.com/ChaoLiAtRIKEN/quanvi.
Recent work on hallucination detection in large language models has shown that, for a fixed pre-trained model and reasoning task, it is possible to estimate the model's confidence in the correctness of its outputs. Such uncertainty estimates have primarily been used to improve truthfulness by detecting or filtering confabulations. In this work, we ask whether these signals can instead be used more proactively to directly improve the accuracy of model-generated answers. We propose USteer, a simple, training-free steering mechanism that adjusts a model's layer-wise activations during inference using the gradient of a confidence measure with respect to the activations. This procedure nudges generation toward outputs with lower uncertainty at inference time, without modifying model parameters or requiring additional supervision. We show that this approach consistently reduces hallucination across a range of reasoning tasks, demonstrating that confidence signals can be leveraged not only for detection, but also for effective inference-time control of model behavior.
Protein Language Models (PLMs) have made remarkable progress following scaling laws established in natural language processing across sequence- and structure-based tasks, yet the potential of tokenization remains underexploited. Unlike human language, proteins preserve structure despite extensive sequence variation, a property standard tokenization strategies fundamentally fail to capture. We introduce ZEST (Zoned Encoding of Sequence Traits), an evolution-informed vocabulary derived from conserved regions of multiple sequence alignments. ZEST allows embedding domain-level biological priors directly at the tokenization stage rather than learning them implicitly through scale. ZEST natively compresses sequences to an average token length of 3.3 residues, enabling our model to process $\geq3380$ residues within a standard 1024-token context window. Building on this, we present LEMON (Layered Extraction of Molecular Ordering from Nature), a compact 200M-parameter sequence-based model for detection of remote homology between protein sequences trained on a single H100 GPU for one week. Despite its modest size, LEMON outperforms models of 600M to 3B parameters and is on-par with TM-Vec. Our results demonstrate that evolution-informed tokenization can substitute for massive parameter scaling, opening a new direction for efficient, biologically-grounded protein representation learning. All code and model weights are publicly available (at \href{https://github.com/LDWLab/ZEST}{github.com/LDWLab/ZEST}) under the MIT license.
Scaling test-time computation in continuous latent spaces offers an attractive alternative to verbose token-by-token generation for multi-step deductive reasoning. However, unregularized recurrent latent updates suffer from a deliberation drift cliff when rolled out beyond training horizons, while strictly energy-conserving or symmetric message-passing dynamics cannot break parent-to-child causal symmetry or suppress cyclic modus-ponens amplification on cyclic distraction graphs. We introduce Typed Edge-Tensor Non-Reciprocal Port-Hamiltonian Deliberation (TE-BPHD), a neuro-symbolic continuous-time latent reasoning architecture that couples skew-symmetric port-Hamiltonian conservation with relation-typed non-reciprocal directional transport and monotone dissipative damping on a semantic entity-state manifold. We prove (i) unconditional input-to-state stability and exponential contraction of autonomous rollouts across arbitrary inference depths without external norm projection, (ii) strict directional DAG reachability separation over symmetric message passing, and (iii) exact cycle-suppression margins under finite cyclic horn perturbations. Across counterfactual open-world deductive benchmarks (PrOntoQA-OOD up to depth 15, ProofWriter-OWA Cyclic at depths 5-10, FOLIO-Symbolic, and Cluttr-Relational), TE-BPHD eliminates the deep-horizon drift cliff, improves depth-10 cyclic reasoning accuracy to 90.0% (+/- 0.8%) over symmetric port-Hamiltonian and standard Transformer baselines, and supports zero-retraining adaptive Lyapunov halting that reduces average solver evaluations by 44% with negligible accuracy loss.
Recent Recommendation Agents (RecAgents) offer a promising alternative by shifting recommendation to an active, user-side paradigm, where generative agents autonomously perceive web platforms, reason over user preferences, and execute decisions. However, existing RecAgents still suffer from two critical limitations: brittle item perception based on noisy item pages, and inefficient long-context reasoning over extended user histories and multi-step interaction traces. To address these challenges, we propose a recommendation-agent framework, termed ReMem, that combines OCR-based multimodal perception with time-evolving dynamic memory. Instead of parsing raw HTML, ReMem observes item-page screenshots and uses an OCR tool to produce structured textual evidence for the reasoning backbone. To support long-horizon preference modeling, ReMem further introduces a chunk-wise sequential update strategy that maintains a bounded textual memory while processing histories longer than the backbone context window; with fixed chunk and memory budgets, the number of update calls grows linearly with history length. This design allows the agent to preserve evolving user preferences without relying on external memory modules or disrupting the standard autoregressive generation process. To train the dynamic memory updater, we further develop a multi-memory GRPO variant whose reward combines recommendation-specific information with final-task metrics. Experiments on three Amazon-derived datasets and a controlled post-retrieval protocol show an average relative improvement of 7.58% over the strongest reported baseline across searching, ranking, and judging tasks.
Noise robustness in automated phonocardiogram (PCG) murmur detection, and how it is measured, remains underexamined despite growing interest in low-resource screening. We evaluate two independently reimplemented pipelines, Hierarchical Multi-Scale Convolutional Network (HMS-Net)--CNN, and Bidirectional Long Short-Term Memory (BiLSTM)--LSTM, under controlled, multi-severity noise with noise-augmented fine-tuning and held-out generalization testing. Under matched aggregation, the complete BiLSTM pipeline outperforms the complete HMS-Net pipeline across all conditions in accuracy and Weighted Accuracy. A stable aggregate accuracy score can misrepresent what individual predictions show: HMS-Net's native aggregation degrades under salt-and-pepper noise far less than majority-vote (MV) aggregation at the same severity, a gap reflecting window-level disagreement its native rule absorbs, while BiLSTM's MV accuracy rises after noise-augmented training even though its individual predictions do not improve. HMS-Net's training effect is significant under one accuracy metric but not another. Noise-robustness conclusions can depend as much on evaluation choices as on the models themselves.
Efficient patient flow coordination across autonomous hospital departments is critical for mitigating overcrowding and balancing resource utilization. While classical queueing theory, specifically open Baskett--Chandy--Muntz--Palacios (BCMP) networks, provides an interpretable mathematical topology for healthcare operations, analytical models rely on stationary assumptions and fixed routing matrices that degrade under state-dependent real-world dynamics. Conversely, centralized reinforcement learning approaches struggle to accommodate the decentralized structure of hospital governance, where individual clinical departments function with localized observations, heterogeneous resources, and divergent operational objectives. In this paper, we present a Multi-Agent Systems (MAS) framework titled \emph{Physics-Informed Multi-Agent Coordination}, which embeds empirically calibrated BCMP queueing topologies as physical priors within a decentralized multi-agent reinforcement learning architecture. Formulated as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP) under coupled resource constraints, our method enables autonomous departmental agents to cooperatively negotiate patient routing and dynamic service scaling. To mitigate environmental non-stationarity without inducing excessive communication overhead, agents exchange localized action fingerprints along network edges and optimize a spatially decomposed reward structure. Empirical evaluations driven by real-world MIMIC-IV patient trajectories indicate that this cooperative multi-agent approach substantially reduces cumulative system delay compared to static Markovian approximations, heuristic dispatching, and independent multi-agent baselines, while maintaining clinical safety constraints.
Localizing AI-edited regions is essential for interpretable forensic analysis, but remains challenging due to subtle and spatially distributed artifacts that are misaligned with semantic or object boundaries. Existing approaches rely on pixel-level supervision from controlled editing pipelines, which is difficult to scale and can introduce misleading signals: artifacts frequently extend beyond annotated regions, while out-of-mask pixels are treated as authentic. This limits models' ability to capture transferable evidence and generalize across generators and datasets. To address these issues, we propose ReGFLoW, a Reconstruction-Guided Fake Localization framework under Weak supervision, which is the first weakly supervised approach for diffusion-edited fake region localization. ReGFLoW requires only real/fake labels at the image-level and uses diffusion reconstruction errors as dense spatial guidance to inject them into both feature and score spaces. Furthermore, by artifact-centric multiple instance learning, ReGFLoW utilizes localized diffusion evidence without relying on semantic-affinity or boundary-based pseudo-mask priors. Extensive experiments show competitive cross-generator localization, while ReGFLoW outperforms all evaluated fully supervised baselines when evaluation includes both partially edited and fully synthetic images and in cross-dataset tests, without target-domain adaptation.
In urban low-altitude flight, buildings reshape ambient wind into spatially varying 3D flow, making unmanned aerial vehicle (UAV) energy depend on local wind exposure as well as path length. However, building-resolved wind is rarely available when a mission must be planned. Computational fluid dynamics (CFD) produces high-fidelity urban flow fields, but each simulation is tied to a fixed inflow boundary condition and can take hours to days, which is incompatible with urban UAV missions that typically last minutes to tens of minutes. We present GeoWind2Plan, a geometry-to-wind-to-planning framework for mission-time 3D urban wind prediction and energy-efficient UAV planning. Given only a background wind vector, 3D building geometry, and a start-goal pair, GeoWind2Plan transforms the building geometry into a reference-wind frame, predicts mission-relevant 3D wind patches with a localized geometry-conditioned neural operator, stitches them into a local wind field, and optimizes a feasible 3D path and speed profile using a physically grounded UAV energy model. Rather than pursuing CFD-perfect reconstruction, GeoWind2Plan targets decision-useful wind prediction: trajectories are planned with predicted wind and evaluated under high-fidelity CFD wind. Across held-out urban domains, wind speeds, and mission wind-angle regimes, GeoWind2Plan performs corridor-localized wind inference in about 3 s, compared with roughly 8 h for CFD. Under CFD evaluation, trajectories planned with GeoWind2Plan reduce energy by 6.9%, 12.7%, and 4.5% in tailwind, headwind, and crosswind missions relative to wind-agnostic planning, recovering 87.9%, 85.7%, and 75.0% of CFD-reference savings. These results show that fast, corridor-localized 3D urban wind prediction can make wind-aware UAV energy planning practical at mission time. Code/dataset: https://github.com/DUAL-Xiao/GeoWind2Plan/ . Demo: https://geowind2plan.com .
Multi-agent large language model (LLM) debate is often evaluated by whether final answers improve, but a changed answer is not necessarily a better one: the same discussion can correct an initially wrong majority or collapse an initially correct one. Standard final-accuracy evaluations conflate these opposing transitions. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol records 253 collapses together with the corrections that offset them, and the correction side changes how interventions should be judged. Replay experiments reveal the central tradeoff: freezing every debate prevents 251 collapses but loses 804 corrections under equal weights, and a leave-one-model-out probe-gated freeze that is tuned for net utility never fires, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman $\rho$ = 0.893, exact two-sided p = 0.0123), but initial-majority accuracy is a close comparator ($\rho$ = -0.821; family partial $\rho$ = 0.767, p = 0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both collapses and corrections. Reasoning-mode models, including one model at two reasoning-effort levels, also show both collapses and corrections. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model rows can be compared under the same denominators and signed utility ledger.
Large language models (LLMs) have made misinformation inexpensive to produce but not to verify, creating a growing asymmetry in the information ecosystem. Under tight time, labor, and budget constraints, media organizations, platforms, and fact-checkers rely on screening to prioritize which content to verify. We introduce VEX-Bench, a unified benchmark for evaluating the verification complexity of LLM-generated misinformation, as perceived during screening, across models and generation methods. Verification complexity is assessed along multiple dimensions derived from journalistic and fact-checking practices, capturing checkability, harm potential, source credibility signals, imposter legitimacy, and expected verification effort. We define the VEX score as an integrated measure combining elicitation yield and verification complexity to quantify how generated content consumes limited verification capacity. We construct a benchmark spanning two misinformation categories, 6 high-stakes domains, and 60 real-world topics, and evaluate 7 frontier LLMs and 7 generation methods, yielding 5{,}880 articles. We employ an LLM-as-judge for scalable evaluation and validate it using content-analysis methodology, including ordinal Krippendorff $\alpha$ for inter-annotator reliability, complemented by fact-checking agents for verification. Our findings show that no single method dominates all dimensions, underscoring the need for multi-dimensional evaluation. LLMs can generate high-VEX misinformation at 3$\times$ to 169$\times$ lower cost than agent-based verification. Such content is often prioritized during screening, consuming scarce verification resources and introducing a systematic risk of misallocation in resource-constrained verification systems. The code is publicly available in our \href{https://github.com/HanxunH/VEX-Bench}{GitHub repository}.
Clinical CT and UHRCT cannot resolve individual trabeculae, whereas synchrotron radiation microCT (SRuCT) provides high-resolution references but is not applicable for in vivo imaging. The two domains differ by a 32x resolution gap, are only coarsely paired, and exhibit severe physical differences including partial volume effects, noise, and artifacts. Existing super-resolution networks and pretrained-prior methods (GLEAN/StyleGAN2, Stable SR/LDM) underperform because they target pixel generation---diverse details and SSIM/PSNR---and do not model these physical differences. Pixel generation for a 32x resolution gap is intrinsically ill-posed. We propose a paradigm shift from pixel generation to topological inference: deterministically predicting invariant microstructures from macro-scale low-resolution inputs, evaluated by morphological parameters. The core of our 2D morphology learning lies in training on 2D slices while evaluating on 3D morphological parameters, ensuring that the learned representations capture true three-dimensional trabecular topology rather than 2D pixel statistics. We realize this paradigm via structural dual super-resolution, coupling forward physical degradation (micro-to-macro) with inverse structural inference (macro-to-micro) through structural duality constraints. The method is an end-to-end, few-shot, compact structural dual network (SDN), comprising a bidirectional modeling network, a multi-scale structural consistency discriminator, and four structural duality constraints. On the test set, SDN achieves morphological parameters largely consistent with SRuCT across six metrics, with SSIM reaching 0.8. Trained on 3.2 um SSRF data, the model generalizes well to 3.25 um isotropic BSRF data from an independent source, validating cross-source generalization and confirming trustworthy structural inference rather than pixel generation.