arXiv:2604.00748v2 Announce Type: replace Abstract: This paper studies efficient data management and timely information dissemination for real-time monitoring of an N-state Markov process, with the objective of enabling accurate state estimation and reliable actuation decisions. We analyze the real-time reconstruction error and the Age of Incorrect Information (AoII), and derive closed-form expressions for their time-averaged values under several sampling and transmission policies. We then formulate and solve constrained optimization problems to minimize the time-averaged reconstruction error and the average AoII under a time-averaged sampling frequency constraint. The resulting optimal sampling and transmission policies are compared to identify the conditions under which each policy is most effective. We further show that directly using the reconstructed state for actuation can degrade system performance, especially when the receiver is uncertain about the state estimate or when actuation is costly. These findings reveal that accurate state estimation alone does not necessarily lead to effective actuation, highlighting the importance of incorporating uncertainty into the decision-making process. To address this issue, we introduce a cost function, termed the Cost of Actions under Uncertainty (CoAU), which characterizes correct and incorrect actuation decisions under receiver-side uncertainty. We propose a randomized actuation policy and derive a closed-form expression for the probability of a correct actuation decision, defined as the event in which the CoAU equals zero. Finally, we formulate an optimization problem to find the optimal randomized actuation policy that maximizes this probability. The results show that the resulting policy substantially reduces incorrect actuator actions.
Science Journals
arXiv:2506.18414v3 Announce Type: replace Abstract: Melanoma is a highly aggressive skin cancer, making early and accurate diagnosis critical. While deep learning excels in skin lesion classification, standard ``black-box" models struggle to explain diagnostic uncertainty, limiting clinical trust. This work introduces a hybrid framework combining a class-aware adversarial Variational Autoencoder and an XGBoost classifier, transcending simple binary classification by leveraging a generative latent space for interpretable decision support. Guided by adversarial training, the model learns the visual characteristics of skin lesions and projects them into a continuous latent space, ensuring that similar images are grouped closely together. Trained on this latent space, the XGBoost classifier achieves a robust AUC of 0.868, competing closely with state-of-the-art models. For borderline cases, the framework enables clinicians to leverage the latent topology through Content-Based Image Retrieval. This provides a dual benefit: it allows the clinician to visually compare an ambiguous lesion against biopsy-confirmed precedents and acts as an early warning sign since a borderline classification can indicate that a lesion shares features of both nevi and melanomas, potentially requiring close monitoring. Our approach translates algorithmic hesitation into transparent, evidence-based visual support, bridging the gap between predictive performance and clinical trust.
arXiv:2604.09698v2 Announce Type: replace Abstract: The growing ubiquity of Extended Reality (XR) is driving Conversational Recommendation Systems (CRS) toward visually immersive experiences. We formalize this paradigm as Immersive CRS (ICRS), where recommended items are highlighted directly in the user's scene-based visual environment and augmented with in-situ labels. While item recommendation has been widely studied, the problem of how to select and evaluate which information to present as immersive labels remains an open problem. To this end, we introduce a principled categorization of information needs into explicit intent satisfaction and proactive information needs and use these to define novel evaluation metrics for item label selection. We benchmark IR-, LLM-, and VLM-based methods across three datasets and ICRS scenarios: fashion, movie recommendation, and retail shopping. Our evaluation reveals three important limitations of existing methods: (1) they fail to leverage scenario-specific information modalities (e.g., visual cues for fashion, meta-data for retail), (2) they present redundant information that is visually inferable, and (3) they poorly anticipate users' proactive information needs from explicit dialogue alone. In summary, this work provides both a novel evaluation paradigm for in-situ item labeling in ICRS and highlights key challenges for future work.
arXiv:2601.23075v2 Announce Type: replace Abstract: On-policy Reinforcement Learning (RL) remains a dominant paradigm for continuous control, yet standard implementations rely on Gaussian actors and relatively shallow MLP policies, often leading to brittle optimization when gradients are noisy, and policy updates must be conservative. In this paper, we revisit actor policy representation as a first-class design choice for on-policy RL. We study discretized categorical actors, which represent each action dimension as a distribution over discrete bins and induce a policy objective analogous to classification cross-entropy loss. Building on architectural advances from supervised learning, we further pair discretized categorical actors with regularized networks, yielding RN-D. Across diverse continuous-control benchmarks, we show that simply replacing the standard Gaussian actor with our proposed actor substantially improves performance, achieving state-of-the-art results within on-policy RL. We release our code at https://github.com/alwaysbyx/RND-RL.
arXiv:2604.21196v2 Announce Type: replace Abstract: We present a comprehensive statistical study of the radial evolution of solar wind turbulence near Mercury's orbit using long-term magnetic field measurements from the MESSENGER mission. Owing to Mercury's highly elliptical orbit and the spacecraft's repeated, extended residence in the upstream solar wind, the data set provides more than 17,000 hours of observations, enabling robust statistics across well-defined heliocentric distance intervals (0.31-0.47 au). We find that inertial-range spectral slopes remain close to -3/2 throughout Mercury's orbit, showing no significant radial evolution. Combined with low magnetic compressibility, this result indicates a stable, predominantly Alfvenic inertial-range cascade already established here. In contrast, kinetic-range spectral slopes exhibit clear radial evolution, becoming progressively shallower with increasing heliocentric distance, highlighting the greater sensitivity of kinetic-scale turbulence to heliocentric conditions. The ion-scale spectral break frequency decreases with distance in the spacecraft frame, while its normalized form increases relative to the local proton cyclotron frequency, demonstrating that the break is not tied to proton cyclotron frequency but reflects evolving local plasma conditions. Magnetic compressibility shows a similar frequency dependence at all distances, with a subtle radial enhancement of compressive fluctuations at kinetic scales. Autocorrelation analysis reveals strong anisotropy, with the correlation times of field-aligned magnetic fluctuations increasing with heliocentric distance, while those of perpendicular fluctuations remain shorter and nearly invariant. Together, these results demonstrate a clear scale-dependent radial evolution of solar wind turbulence near Mercury-s orbit, providing new constraints on the development of kinetic processes in the inner heliosphere.
arXiv:2604.23178v2 Announce Type: replace Abstract: LLM-as-a-Judge has become the dominant paradigm for evaluating language model outputs, yet LLM judges exhibit systematic biases that compromise evaluation reliability. We present a comprehensive empirical study comparing nine debiasing strategies across five judge models from four provider families (Google, Anthropic, OpenAI, Meta), three benchmarks (MT-Bench n=400, LLMBar n=200, custom n=375), and four bias types. Our headline practical finding is that a mid-tier model with the right debiasing can outperform frontier judges at a fraction of the cost: Gemini 2.5 Flash with the Combined Budget strategy reaches the highest agreement of any configuration we tested (71.0%, kappa=0.549) at ~$0.001 per evaluation, about 15x cheaper than the best frontier setup (Claude Sonnet 4, 69.5%, ~$0.015). Other key findings: (1) Style bias is the dominant bias (0.10-0.76 across models, favoring markdown over plain prose), far exceeding position bias (<=0.04), yet is rarely studied. (2) Verbosity bias is heterogeneous when measured length-aware: Pro, Flash, and Llama prefer longer answers (+0.24 to +0.44), Claude prefers concise (-0.12), and GPT-4o is neutral (-0.04); on truncation controls all models correctly prefer the complete response (0.88-1.00 accuracy). (3) Debiasing helps multiple models: Claude S8 (+11.5 pp), Flash S8 (+7.5 pp), and Claude S5 (+7.3 pp) survive Holm-Bonferroni correction, with Flash S1 (+4.7 pp) and Llama S8 (+4.5 pp) also significant. We release our evaluation framework, the 375-pair controlled dataset, and per-instance cached results for all nine strategies.
arXiv:2606.25360v1 Announce Type: new Abstract: While end-to-end Vision-Language-Action (VLA) models show promise in robotic manipulation, their monolithic paradigm inherently couples semantic reasoning and spatial control. This creates a severe alignment bottleneck, limiting precise target disambiguation in data-constrained imitation learning. To overcome this, we propose SVP-IL, a decoupled architecture that explicitly extracts spatial visual grounding from the action generation loop. By leveraging vision-language foundation models, we parse instructions into zero-shot geometric masks, translating language into explicit Spatial Visual Prompts (SVP). These priors are injected into a continuous action generator via a lightweight direct feature-level fusion mechanism. This integration provides explicit and uncorrupted spatial gradient guidance while ensuring highly stable optimization under low-data regimes. Extensive experiments demonstrate that SVP-IL significantly outperforms state-of-the-art VLAs and pure visuomotor baselines. Trained on as few as 50 to 100 demonstrations, SVP-IL improves average success rates on highly ambiguous language-conditioned tasks from 24.0% to 39.5%, achieving 67.8% on standard benchmarks. Real-world robotic experiments further validate its robustness and data efficiency in unstructured physical environments.
arXiv:2606.25420v1 Announce Type: new Abstract: In this work, to address the challenge of joint channel estimation and data detection (JED) for orthogonal delay-Doppler (DD) division multiplexing (ODDM) in doubly selective channels, we propose an orthogonal approximate message passing (OAMP)-aided JED (OAMP-JED) receiver. We first formulate a bilinear cross-domain JED model, which can be linearized into separate channel estimation and data detection subproblems. The proposed OAMP-JED receiver alternately executes two OAMP modules for these subproblems, effectively coupled through a variational noise term to account for model uncertainty. Leveraging OAMP's error orthogonality, we derive closed-form scalar-variance updates to enable efficient and principled soft information exchange between the modules, thereby mitigating error propagation during JED. Simulation results show that, for both uncoded and coded ODDM, OAMP-JED achieves a lower bit error rate (BER) than benchmark schemes. Moreover, its BER performance closely approaches that of OAMP with perfect CSI.
arXiv:2606.25398v1 Announce Type: new Abstract: Reward design remains a major bottleneck in reinforcement learning for robot locomotion, where successful policies often depend on carefully tuned, task-specific reward functions. Preference-based reinforcement learning offers an alternative, but existing LLM-based methods typically ask for a single overall judgment between behaviors, making it difficult to capture the multiple competing objectives that underlie high-quality locomotion. We present Multi-Objective AI-Informed Preference Learning (MAPL), a framework that learns locomotion rewards from high-level natural language objectives rather than manually engineered reward equations. MAPL prompts a large language model to compare trajectories independently along semantically meaningful criteria, using generic language descriptions that are terrain-invariant and require little domain expertise. These objective-wise preferences are used to train a multi-head preference scoring model, whose outputs are aggregated to form a scalar reward for policy optimization. Across four quadruped locomotion environments, MAPL trains policies using only LLM-generated preferences and achieves performance comparable to or better than expert-designed rewards, while eliminating task-specific reward engineering.
arXiv:2510.21700v2 Announce Type: replace Abstract: In this paper we construct distance sketches for intersection graphs of arbitrary path-connected regions in the plane (known as the string graphs) in the constant and $1+\varepsilon$ distortion regimes. Furthermore, the distance sketches themselves are planar graphs. First, we show that every unweighted string graph $G$ has an $O(1)$-distortion planar emulator: that is, there exists an edge-weighted planar graph $H$ containing every vertex in $G$, such that every pair of vertices $(u,v)$ satisfies $\delta_G(u,v) \le \delta_H(u,v) \le O(1) \cdot \delta_G(u,v)$. Furthermore, we show that for any constant $\varepsilon > 0$, there is an edge-weighted planar graph $H'$ such that every pair of vertices $(u,v)$ satisfies $\delta_G(u,v) \le \delta_{H'}(u,v) \le (1+\varepsilon) \cdot \delta_G(u,v) + O(\varepsilon^{-4}\textrm{poly}\log n)$. No previous constructions of sparse distance sketches were known even for intersection graphs of simple shapes like axis-parallel rectangles or fat convex polygons. As applications, we construct the first $(1+\varepsilon, +O(1))$ mixed-distortion tree cover and distance oracle for arbitrary string graphs, as well as the first additive $+(\varepsilon\Delta+O(1))$-distortion embedding of string graphs $G$ with diameter $\Delta$ into graphs of constant treewidth $O(\varepsilon^{-4})$.
arXiv:2605.02558v3 Announce Type: replace Abstract: Accurate badminton stroke prediction is crucial for fine-grained sports analysis and tactical decision support. However, existing methods struggle to model rich temporal context. This paper introduces TemPose-TF-ASF (Adjacent-Stroke Fusion), a context-aware extension of TemPose. It enhances stroke recognition by incorporating stroke-type information from both preceding and subsequent strokes. A two-stage training and inference strategy is adopted. Preliminary predictions from the baseline model are reused as estimated temporal context. These predictions guide the joint optimization of the ASF module and the classifier. By explicitly modeling bidirectional temporal stroke dependencies, the proposed method can be seamlessly integrated into existing state-of-the-art models. Experiments on a large-scale badminton match dataset show consistent improvements over the baseline and its variants in terms of Accuracy and Macro-F1. Moreover, integrating ASF into other advanced methods yields notable performance gains. These results demonstrate strong transferability and generalization capability.
arXiv:2605.04335v2 Announce Type: replace Abstract: Computational fluid dynamics and fluid-structure interaction simulations involving moving and deforming bodies is extremely hard. In this work, we present a graphical processing unit (GPU) optimized implementation of the sharp-interface immersed boundary method. The method allows performing simulation around complex stationary as well as moving bodies on a Cartesian grid. We base our implementation on the ViCar3D framework and make use of OpenACC, CUDA, NCCL and MPI. We test the implementation across grid sizes ranging from O(10million) to O(1billion) points and achieved a 20X speedup compared to existing CPU implementation. We next present our multi-GPU implementation by utilizing CUDA streams and NCCL communicators. This enables us to obtain a >90% strong and weak scaling efficiencies. Next we demonstrate the capability of the developed software to simulate a turbulent fluid flow and coupled fluid-structure interaction in flapping bat wing in flight at Re=5000.
arXiv:2605.05097v3 Announce Type: replace Abstract: LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and one slow, following the Benna-Fusi model of synaptic consolidation. From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting are expected to emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics. This workshop article describes an early-stage conceptual design without experimental evaluation.
arXiv:2606.25447v1 Announce Type: new Abstract: Tool-integrated LLM agents are often wrapped within a harness: the scaffolding that determines which tools are exposed, how they are described, and what auxiliary information accompanies each per-step observation. While agents are routinely post-trained, this scaffolding is typically treated as a fixed engineering detail, with design effort limited to the training-free regime. Moreover, existing post-training algorithms assume a static environment, even though tool environments and tasks often shift upon deployment. To address this gap, we extend $\texttt{ALFWorld}$ (i) to treat the harness as a controllable design dimension and (ii) to support evaluation under task and tool environment shifts. Building on this, we systematically analyze how the harness design influences post-training in both in-distribution and out-of-distribution (OOD) settings. We empirically show that harness-aware post-training not only improves in-distribution performance but also enables agents to robustly adapt to OOD settings. Under a harness with minimal design effort, post-training suffers a drastic performance drop under stronger tool environment shifts, further highlighting the importance of harness-aware post-training under such shifts.
arXiv:2606.25448v1 Announce Type: new Abstract: We show that there is a simple syntactically-defined subclass of higher-order benchmarks in the termination problem database for which rewriting according to Nipkow's higher-order rewrite systems (HRSs) and rewriting according to a beta-first strategy in the semantics of TermCOMP's higher-order category coincide. This lays the formal foundation for an HRS (sub)category in TermCOMP which would allow more tools to compete against each other.
arXiv:2606.25453v1 Announce Type: new Abstract: Modern GPUs devote an increasing silicon budget to low-precision matrix-multiplication units, widening the precision-throughput gap for scientific computing workloads. Ozaki Schemes I and II offer an alternative by reconstructing high-precision general matrix multiplication (GEMM) from low-precision operations, yet existing implementations leave substantial performance untapped. In particular, intermediate results are repeatedly materialized in global memory, making data movement the dominant bottleneck. We present EmuGEMM, fused integer Tensor Core kernels for NVIDIA Hopper and Blackwell GPUs that eliminate redundant memory round-trips in both Ozaki schemes. Using Scheme I, EmuGEMM sustains up to 1,639 Top/s on Hopper (83% of INT8 peak) and 3,654 Top/s on Blackwell (81%). For large matrices, EmuGEMM surpasses cuBLAS TF32 throughput by up to 1.4x on Hopper and 1.7x on Blackwell, at comparable accuracy. Using Scheme II, EmuGEMM extends to complex arithmetic and outperforms cuBLAS ZGEMM by up to 2.3x on Hopper and 5.5x on Blackwell.
arXiv:2606.25454v1 Announce Type: new Abstract: In this study, the learning-based inertial odometry problem is investigated using raw IMU measurements obtained from the EuRoC MAV benchmark dataset. Instead of absolute position regression-a formulation that may lead to large constant errors-the models are trained to estimate the incremental displacement ({\Delta}p) over a fixed 50 ms sliding window, and the full trajectory is reconstructed through numerical integration. A standard Multi-Layer Perceptron (MLP) is compared with a Kolmogorov-Arnold Network (KAN) equipped with learnable B-spline activations. Although KAN has 6.9 times fewer parameters than MLP (8,444 versus 57,859), it produces a 44% lower error in terms of final cumulative drift on the test trajectory (9.61 m versus 17.23 m). In addition, KAN exhibits more stable behavior in terms of long-term error accumulation, with lower P_50 and P_90 cumulative drift values. These findings indicate that learnable B-spline-based activations have the potential to reduce error accumulation in the inertial odometry problem.
arXiv:2606.25530v1 Announce Type: new Abstract: Software performance optimization is a notoriously complex and manual task. Despite the growing use of Large Language Models (LLMs) for code refinement, we still lack benchmarks that capture how optimization actually happens in real-world codebases. Existing frameworks often oversimplify the problem by focusing on isolated functions or a single performance metric, missing the critical trade-offs between execution time and memory footprint, the inherent noise of the measurement environment, and the variability introduced by different input data and execution conditions. We address this by introducing SWE-Pro, a repository-level benchmark derived from 102 expert-written optimizations from open-source projects. Unlike previous benchmarks, SWE-Pro pairs each task with parameterized tests to evaluate runtime, peak memory, and Time-Weighted Memory Usage (TWMU) across varying input data and execution conditions under noise-aware measurement conditions. Our evaluation shows that current LLMs struggle significantly: runtime gains are negligible, and memory optimizations are nearly non-existent. This stands in sharp contrast to expert implementations, which achieve an aggregate speedup of 15.5x and peak memory reduction of 171.3x over benchmark tasks. Expert-written improvements are observed in 91.2% of tasks for runtime and 65.7% for peak memory. Our findings expose a substantial gap between current LLM capabilities and the demands of expert-level engineering.
arXiv:2605.05486v2 Announce Type: replace Abstract: We demonstrate that the conversion gain of a superconducting hot-electron bolometer (HEB) mixer can be increased by biasing the device within the negative differential resistance (NDR) region of its current-voltage characteristic. Although NDR biasing has historically been avoided due to MHz-range resistive oscillations, we show that these oscillations arise from an LC resonance formed by the bias-T inductance and the effective thermal capacitance of the HEB. By applying stability criteria analogous to those developed for tunnel diodes, we redesigned the embedding circuit to suppress this resonance and achieve stable NDR operation. Direct measurements using two monochromatic 2.5-THz sources confirm the predicted gain enhancement. These results establish NDR biasing as a viable method for improving HEB mixer performance and motivate further studies of noise behavior and circuit optimization.
arXiv:2605.12767v2 Announce Type: replace Abstract: Radioactive molecules provide a powerful new platform in the search for new physics at energy scales complementary to high-energy particle colliders. By combining enhancements from nuclear properties with the sensitivity and control offered by molecular structure, experiments with radioactive molecules offer great reach in the search for new physics beyond the Standard Model. Rapid progress in this field is being driven by advances in the production and control of radioactive molecules, alongside the development of new experimental tools and theoretical techniques. In this Perspective, we discuss the current status and future prospects of this rapidly developing, interdisciplinary field at the intersection of nuclear physics, atomic and molecular physics, and particle physics.
arXiv:2605.26845v2 Announce Type: replace Abstract: The growing demand for efficient communication in distributed training and inference has sparked significant interest in reconfigurable photonic interconnects across both academia and industry. Mixture-of-Experts (MoE) models, with their highly skewed communication patterns, present a natural opportunity for such circuit-switched fabrics. However, existing approaches largely optimize communication in isolation, overlooking the interaction between communication and the expert computation that follows. In this paper, we revisit circuit scheduling for all-to-all communication in MoE execution. We show that the dispatch--compute--combine structure fundamentally challenges classical scheduling techniques such as Birkhoff--von Neumann (BvN) decomposition. First, MoE communication matrices are rarely doubly stochastic, introducing significant scheduling bubbles in BvN-based schedules. Second, while decomposition enables communication--compute overlap, the excessive number of matchings produced by BvN fragments execution into small batches, leading to severe compute inefficiencies due to fixed execution overheads. Motivated by these observations, we explore a simple greedy max-weight decomposition strategy that bounds the number of matchings while preserving large batch sizes per matching. Despite its simplicity, the approach significantly improves overlap efficiency, reduces compute overheads, and approaches the performance of an ideal congestion-free all-to-all.
arXiv:2605.29129v2 Announce Type: replace Abstract: Agentic AI systems are increasingly being explored as production infrastructure: they reason over multiple steps, call tools, act through workflows, and adapt through memory and feedback. These systems create governance challenges that are not fully captured by traditional software or predictive ML technical debt. We define Agentic Technical Debt as the accumulated liability created when prompts, memory, tool schemas, orchestration graphs, control policies, and observability routines are patched together faster than they can be validated, standardized, and governed. We define Stochastic Tax as the recurring operating burden of keeping probabilistic agent behavior within acceptable bounds. The distinction matters: debt is a stock of design and governance liability, while the tax is a flow of operating cost that arises because stochastic agents act through tools and workflows. We outline how managers can make both visible through lightweight dashboards and governance controls.
arXiv:2605.29682v2 Announce Type: replace Abstract: Agent harnesses shape language-model performance by controlling tool use, feedback, verification, memory, and repair. Yet raw test-time expenditure, such as tokens, tool calls, wall time, or cost, cannot distinguish useful feedback from redundant or unstable interaction. We introduce \emph{Effective Feedback Compute} (EFC), a trace-level scaling coordinate for informative, valid, non-redundant, and retained feedback. We further define Estimated-EFC, NRS-EFC, harness efficiency $\eta$, and task-demand normalization for realistic traces and heterogeneous tasks. Across synthetic, real, held-out, and prospective evaluations, EFC-based coordinates outperform raw-compute baselines and SAS. Oracle-EFC/$D_{\mathrm{task}}$ reaches $R^2=0.99$ in controlled scaling, and NRS-EFC/$D_{\mathrm{task}}$ reaches $R^2=0.93$ on real traces where raw compute has near-zero or negative fit. Finally, \ours uses EFC as a companion control layer for existing harnesses, improving mean pass rate from $61.2\%$ to $68.2\%$ while reducing mean raw cost from $213.8$ to $85.1$ under matched settings. These results suggest that harness scaling depends on durable, task-sufficient feedback rather than raw computation alone.
arXiv:2511.22411v2 Announce Type: replace Abstract: 3D head stylization enables expressive reimagining of human faces for creative visual experiences in digital media. Existing 3D-aware methods often require computationally intensive optimization or per-style fine-tuning, limiting flexibility and user control. To overcome these challenges, we introduce StyleFusion360, a diffusion-based framework for multi-view consistent, identity-preserving 3D head stylization from a single style reference image, without per-style training. Our approach enhances the Style Fusion Attention mechanism with a style-conditioned key modulation mechanism that aligns content and style representations for fine-grained and controllable stylization. We further provide a user-controllable slider for adjusting stylization intensity. In addition, StyleFusion360 supports local multi-edit stylization, enabling targeted edits such as modifying hair or eyes independently. Extensive experiments on FFHQ and RenderMe360 demonstrate that StyleFusion360 produces high-quality, controllable, and visually compelling stylizations, outperforming state-of-the-art GAN- and diffusion-based methods across diverse style domains.
arXiv:2512.01759v3 Announce Type: replace Abstract: We investigate the potential of weights to serve as effective representations, focusing on neural fields. Our key insight is that constraining the optimization space through a pre-trained base model and low-rank adaptation (LoRA) can induce structure in weight space. Across reconstruction, generation, and analysis tasks on 2D and 3D data, we find that multiplicative LoRA weights achieve high representation quality while exhibiting distinctiveness and semantic structure. When used with latent diffusion models, multiplicative LoRA weights enable higher-quality generation than existing weight-space methods.