Papers

Empathetic spoken dialogue requires models to use both what is said and how it is said to decide how to respond. Explicit CoT can improve paralinguistic perception and make acoustic cues more explicit in replies, yet does not ensure their effective use in response planning. We call this mismatch the perception-reasoning gap. In addition, CoT may not fully capture acoustic cues in words, and generating it adds inference latency. To address these limitations, we introduce LoopSLM, which builds on looped Transformers for latent reasoning, reusing a decoder block to refine hidden states with acoustic grounding at every pass. Its two-stage training further narrows the perception-reasoning gap by separating learning to reason from learning to respond, enabling direct inference without CoT. On EchoMind, LoopSLM improves paralinguistic understanding, reasoning, and reply quality over Qwen2.5-Omni-7B. Against the CoT-SFT baseline, LoopSLM gains over 20 points in reasoning accuracy while generating 64.5% fewer tokens at half the latency. It also outperforms Qwen3-Omni-Thinking on most empathetic reply metrics with 34x lower latency. Despite training only on dialogue data, LoopSLM improves accuracy on general audio benchmarks.

Geometric representation learning predominantly scaffolds representations onto flat Euclidean subspaces or compact product tori ($\mathbb{T}^K$). However, flat manifolds possess vanishing curvature and polynomial volume growth, inherently suffering from metric distortion when embedding multi-scale, tree-like visual hierarchies. While hyperbolic spaces ($\mathbb{H}^m$) circumvent this via constant negative curvature ($K<0$) and exponential volume expansion, prior hyperbolic deep architectures are hindered by computationally cumbersome Riemannian optimization, non-linear gyrovector calculus, and floating-point instabilities. In this work, we introduce Minkowski Attractor Networks (MAN), an operator-splitting-inspired framework that embeds representations within pseudo-Riemannian Minkowski spacetime ($\mathbb{R}^{1,m}$). By framing hyperbolic manifolds as quadric level sets, MAN resolves hyperbolic geometry by combining linear Lorentz group transport with non-linear cone lifting and closed-form radial rescaling, evaluating in a single forward pass without numerical ODE solvers or iterative retractions. We establish MAN-2D ($\mathbb{R}^{1,1} \to \mathbb{H}^1$) as our primary, high-throughput visual backbone, which maximizes channel factorization granularity into $D/2$ independent two-dimensional Minkowski blocks. We further formulate MAN-4D ($\mathbb{R}^{1,3} \to \mathbb{H}^3$) as a spacetime extension, leveraging a commuting Cartan-subalgebra parameterization of $\mathrm{SO}^+(1,3)$ to evaluate 4D Lorentz isometries via two commuting 2D planar maps without matrix-exponential overhead.

Inpainting 3D Gaussian Splatting scenes, a key challenge in 3D editing, requires generating plausible content within a masked region of 3D space. Prior approaches rely on 2D diffusion models to produce one or several inpainted reference views, making them susceptible to challenges associated with multi-view inconsistency and lengthy optimization times. Departing from these approaches, we introduce a reference-free Gaussian splatting inpainting method operating natively in 3D. Our method leverages the embedding space of a large, pre-trained 3D prior, combined with a structure completion network to feed a generative prior which reconstructs the missing region's geometry and appearance. Performing content generation entirely in 3D, it avoids the need to reconcile inconsistencies of multiple inpainted reference images, and is faster than related 2D-based methods. We demonstrate the effectiveness of our method qualitatively and quantitatively, through extensive experiments and a user study. To the best of our knowledge, this work is the first Gaussian splatting inpainting method to operate in the learned representation space of a 3D-native generative prior without relying on inpainted reference views.

Concurrent games and strategies based on event structures are an established and expressive model of interactive computation. In this paper we revisit the problem of equipping this model with symmetry. Symmetry is essential for many applications, but it is also a source of mathematical complexity, not least because strategies with symmetry can be compared up to several distinct notions of equivalence. Our first contribution is a complete characterization of two classes of strategies for which composition admits an identity. For these two classes the identity laws hold, respectively, up to a "weak" and a "strong" notion of equivalence. The main result is that the weak class of strategies corresponds precisely to those inducing Street fibrations over the game. We also show how the established "thin" approach to symmetry also fits in this general framework. A second contribution is a formal connection between games and setoid-valued profunctors. This follows a long tradition of relating games and relational models, but here additionally includes a very general model of symmetries. Our characterization of strategies as Street fibrations makes the connection to profunctors significantly easier to establish.

In this letter, we develop hierarchical Fourier phase projection (HPP) for the calculation of local electronic observables. The method couples the spatial decay of the density matrix with a hierarchy of Fourier probes, improving accuracy through refinement, and making use of all previously evaluated responses. The variation in the calculated observables during the refinement process carries information on the residual error. We use a locality-based model to estimate the error from these changes with no extra probing calculations. In particular, we demonstrate the method for electron densities and nonlocal pseudopotential forces for fixed Kohn-Sham Hamiltonians of semiconductors, metals and thermally perturbed silica. The convergence is significantly accelerated by the spatial decay of the density matrix, especially if the density matrix is short-ranged. The number of probes necessary to achieve a specific local accuracy is only weakly dependent on the system size within the range of consideration. HPP provides a route toward linear-scaling electronic-structure calculations in which the computational effort for local observables can be matched to the required accuracy, offering an error-aware strategy for large-scale first-principles simulations.

Designers often speak while sketching when explaining ideas, yet AI design tools often rely on sketches or prompts, overlooking context expressed as ideas develop. We developed a sketch-based AI design interface that jointly interprets sketches and concurrent speech. Through a between-subjects study ($N=24$), we examined how speaking while sketching steers human–AI design ideation compared with sketches alone. For creativity support, concurrent speech supported natural expression of design intent and efficient visualisation. For human–AI collaboration, speech helped establish a shared understanding of design intent, supported significantly higher perceived alignment ($p<.05$), and enabled participants to guide AI contributions as ideas co-evolved. We discuss how future human–AI design tools could support dynamic alignment, broader multimodal expression, and human–AI co-creativity.

Unconventional superconductivity commonly emerges in systems characterized by strong electronic correlations, with layered copper-oxides serving as a canonical example. Observation of spin-momentum locking within Bi-family of the cuprates, compatible with the presence of non-negligible Rashba-type spin-orbit coupling (RSOC), calls for an investigation of the joint effects of electronic correlations and RSOC on pairing in copper-oxide and related superconductors. Employing statistically-consistent variational approximation (SGA), we carry out such an analysis by constructing the phase diagram for the case of square-lattice t-J-U model incorporating RSOC. We also investigate the effects of time-reversal-symmetry breaking by Zeeman field, as well as characterize emergent topological superconducting (TSC) states. Chern number $C = \pm 4$ TSC is found in a broad regime of on-site Coulomb repulsion close to half-filling. The latter is not governed by correlations and emerges also within the weak-coupling Bogoliubov-de Gennes (BdG) scheme. Yet, we identify a distinct $C = \pm 2$ TSC state that is driven specifically by electronic correlations via a topological transition occurring with no bulk quasiparticle gap closure, and is accompanied by discontinuous metamagnetic and Lifshitz transitions. Moreover, a qualitatively distinct doping evolution of the $d$- and $p$-wave components of the underlying mixed-parity SC order parameter above the metal-to-insulator transition is demonstrated. Our work points toward the relevance of joint correlation and RSOC effects beyond BdG scheme to phase diagrams of RSOC-coupled superconductors.

Solar spicules are ubiquitous and strongly overlap in off-limb observations, which complicates the interpretation of individual events. We aim to infer population-level properties of spicules from IRIS Mg II k off-limb spectra using a forward model that explicitly accounts for line-of-sight superposition and instrumental effects. We constructed a stochastic 3D spicule-population model and synthesized Mg II k spectra with radiative transfer, including instrumental degradation and photon noise. We then constrained the hyperparameters that describe the underlying spicule population with likelihood-free Bayesian inference. This approach combines approximate Bayesian computation and an adaptive population Monte Carlo sampler, comparing observed and simulated datasets through principal-component summary statistics at five heights above the limb. The inferred posterior constrains effective population-level properties of the spicules that dominate the observed Mg II k variability and height dependence. These constraints provide a route to connect off-limb spectroscopy with forthcoming spectropolarimetric diagnostics of the chromosphere–corona interface.

Vision-language-action and world-action models have demonstrated impressive capabilities in robotics, yet generalization to unseen tasks remains challenging. More recently, general-purpose multimodal agents have shown great potential for zero-shot robotic task solving. However, they often incur high execution costs by reasoning and exploring the physical world from scratch. To reduce these costs, we introduce RoboSkill, a framework that connects skill acquisition and reuse through an Explore, Execute, Evolve loop. Within this loop, the agent explores to gather task-relevant information, executes tasks while adapting to feedback, and evolves its skill library based on execution records. It then reuses these skills to guide exploration and execution in the next cycle, closing the loop. To improve loop efficiency, we complement vision with tactile feedback to reduce uncertainty during physical interaction. We further augment textual guidance with reusable code to reduce reasoning overhead during skill reuse. On LIBERO-10, RoboSkill improves first-episode success rates by 12.5–25.0 percentage points and reduces average runtime by 7.6–72.4% across four agents. On real robots, it improves success rates by 8.3 percentage points and reduces average runtime for successful trials by at least 14.4%.

Predicting canopy height from medium-resolution satellite imagery is a common and scalable approach for assessing the condition of the world's forests, which play a crucial role in climate change mitigation. While Transformer-based architectures have shown strong performance in many domains, their straightforward application to dense (i.e., pixel-level) regression tasks often yields suboptimal results. In particular, the patch size has a crucial impact on the model performance. In this work, we consider pixel-level attention schemes and show that the resulting models generally outperform those relying on larger patch sizes. However, pixel-level attention can be a prohibitively resource-intensive operation. For this reason, we conduct an extensive experimental study using efficient attention variants to identify favorable trade-offs between prediction quality and resource requirements, facilitating the practical deployment of the proposed models. In addition, we perform a comprehensive comparison with several well-established models in the field and show that, with suitable hyperparameter choices, Transformer-based architectures can outperform competing approaches. Our findings provide practical guidance for designing models for pixel-level regression tasks on medium-resolution satellite imagery, including canopy height and biomass estimation, soil moisture mapping, and yield forecasting.

Protein optimization aims to discover high-fitness sequences under a limited experimental budget. Existing machine-learning methods use task-specific predictors, biological priors, or ranking-aware objectives to guide which variants are tested in the next experimental round. However, these methods cannot adapt to shifts in the reliability of predictive evidence as measurements accumulate and ensure the correct ranking of key high-fitness candidates. To address these challenges, we propose Batch-Aligned Tail Arbitration (BATA), which uses experimental feedback to adaptively combine prior-informed and task-specific rankings for next-batch selection, with calibration focused on the batch-aligned high-fitness region. Across measured GB1, PABP, and TrpB landscapes, BATA achieves the best mean task rank (1.67) in final best fitness after 480 measurements. Controlled comparisons further show task-dependent gains from high-fitness calibration and batch alignment. Our work introduces feedback-calibrated predictor arbitration, where experimental feedback dynamically determines how predictive evidence guides next-batch selection, opening a new direction for protein optimization.

Studying how fine-tuning shapes refusal and noncompliance behaviour requires knowing which training examples refuse or otherwise fail to fulfil the request. Existing annotations cover evaluation sets, which are far smaller than training corpora. We present CompOrca, compliance labels for all 4,233,923 examples of the OpenOrca corpus. Every example was classified as compliant or noncompliant by five passes of an open-weight LLM judge (LongCat-2.0, 1.6T parameters). The corpus is released as unanimous compliance (94.75%), unanimous noncompliance (1.28%), and nonunanimous rows (3.97%), with the raw vote counts. A single pass flags 2.7-3.2% of the corpus as noncompliant, while only 1.28% is flagged by all five, so the most ambiguous rows can be filtered out. Against 450 human-annotated examples (150 annotated twice; human-human $\kappa=0.93$), the unanimous compliance and noncompliance labels are 97.3% and 86.7% precise. The noncompliance label is a high-precision subset of the corpus's noncompliance. Published refusal-detection methods recall between 0.4% and 94.1% of human-labelled noncompliance. We release the full corpus with its per-row labels and vote counts at https://huggingface.co/datasets/cemiu/CompOrca

We study heavy-heavy-light-light correlators in holographic CFTs when the heavy operator is dual to a compact object localized on scales much smaller than the AdS radius. In the long-wavelength regime, the internal structure of the heavy state is described by a worldline effective field theory, whose local operators encode finite-size response coefficients such as tidal Love numbers. Focusing on weakly self-gravitating objects, we show that the Witten diagrams generated by local heavy-light interactions reorganize into the Born series of an effective wave equation for the light bulk field. We reproduce the same correlator by solving this wave equation with localized interactions and identify how the worldline Wilson coefficients are encoded in the CFT data. Our results provide a direct relation between the finite-size response of compact objects in AdS and the dimensions and OPE coefficients of heavy-light and light-light composite operators in the dual CFT.

We study stochastic systems driven by a Brownian sheet and formulate a Markov property adapted to their two-parameter structure. For time-space homogeneous Itô sheets, the future evolution from a point $(t,x)$ is not determined by the single value $Y(t,x)$ alone, but also by the boundary data along the horizontal and vertical edges issuing from $(t,x)$. We therefore show that the enlarged state consisting of the triple $(Y,D_1Y,D_2Y)$ satisfies a natural Markov property with respect to the past sigma-algebra generated by the two coordinate histories of the Brownian sheet. The result is illustrated by examples showing the strong martingale property for the Brownian sheet and for simple Brownian sheet systems. We also recall a Brownian sheet Girsanov theorem and use it to treat drifted sheets.

Let $p:X\rightarrow\Delta$ be a proper holomorphic submersion from a Kähler manifold $X$. We prove that, if $K_{X_t}$ is nef for every $t\in\Delta$, then $\dim_{\mathbb C}H^0(X_t,mK_{X_t})$ is independent of $t$ for every integer $m\geq1$. This gives a partial answer to Siu's conjecture on the invariance of plurigenera for Kähler families.

In a continuous coordination game, under myopic best response, players change strategies, and thus strategic communities change their shapes in time. We show that the boundaries of these strategic communities evolve according to Free Boundary Mean Curvature Flow in the continuous time limit. The convergence relies on an approximation scheme for the Mean Curvature Flow, which has previously been used to describe other threshold dynamics. With this equivalence, we can characterize Nash equilibria of the continuous coordination game as minimal surfaces and explore related consequences in biased versions of the same game.

Simulating nonlinear dynamics with quantum computers has gained increasing attention. In general, such simulations require additional quantum resources because unitary quantum evolution is linear. A fundamental question is how nonlinear dynamics can be embedded into fully coherent, ancilla-free unitary circuits and how the complexity of the dynamics governs the required quantum resources. In this work, we generalize the ancilla-free double-bracket quantum algorithm for imaginary-time evolution by replacing its state-independent Hamiltonian with a state-dependent Hermitian operator. Our framework recursively calls an initial state preparation oracle and its inverse, and prepares the target solution to any prescribed accuracy using a fully coherent, ancilla-free unitary embedding. We relate the query cost to the complexity of the nonlinear dynamics, specifically their sensitivity to initial conditions. We obtain query upper bounds of $\exp(O(T))$, $\exp(O(T^2))$, and $\exp(\exp(O(T)))$ when the distance between solutions contracts at least exponentially (contractive), does not increase (nonexpansive), or grows at most exponentially (expansive), respectively, where $T$ is the target evolution time. For the discrete Gross–Pitaevskii equation, our ancilla-free double-bracket circuit achieves optimal worst-case query complexity $\Theta(e^{gT/2})$ over a specified family of single-qubit initial states, where $g>0$ is the nonlinearity strength. These results connect the complexity of nonlinear dynamics to the query cost of coherent quantum simulation and provide a foundation for designing ancilla-free unitary embeddings with optimal query complexity.

The ByteTrack algorithm is a widely used and computationally efficient multi-object tracking architecture. Its core innovation lies in the combination of lenient bounding box associations with tracklet similarity matching to robustly deal with object occlusions. However, this strategy is nevertheless vulnerable to erroneous track reclassification and identity switching, as detection confidence scores dictate association priority. To address this, I present a simple enhancement of the ByteTrack architecture–named ByteTraX–that optimises track continuity via a single unified matching threshold, while penalising identity switches through stringent track initiation criteria. This approach achieves consistently improved performance across a range of diverse benchmarks including GMOT-40, LC-MOT, SportsMOT, TeamTrack, DAMUNT, and DeepSea-MOT, while simultaneously increasing processing speed by >10%. Specifically, results demonstrate a >40% reduction in identity switches, accompanied by mean increases in HOTA of 3.6, IDF1 of 5.6, and FPS of 6.3. As such, adoption of the ByteTraX algorithm has the potential to substantially enhance tracking performance over the ByteTrack baseline, while retaining the efficiency needed for real-time deployment. To facilitate usage, I provide the source code, integration functionality for the YOLO family of object detection models, and deployment instructions via an open source repository.

Many recommender services repeatedly encounter cold-start cohorts, where new users arrive with little or no interaction history. This creates two challenges: learning user preferences quickly from limited feedback and sustaining useful recommendations when each user has a finite catalog that can become repetitive or depleted over time. We propose CohortMix-TS, a warm-started mixture bandit that learns latent user groups from earlier cohorts and uses available metadata to construct group-informed priors for new users. Starting from these fixed priors, the model personalizes independently as feedback from each user becomes available. Session slates combine Thompson sampling with diversity and inventory-depletion controls. We evaluate CohortMix-TS through simulation, semi-synthetic experiments, and a 25-day randomized in-the-wild deployment with 713 registered participants in a Campus Games quiz application. Our evaluations show that cross-cohort transfer improves early recommendation quality and user-level regret, while inventory-aware slate construction helps prevent premature exhaustion of preferred items. In the field deployment, treatment users also showed a larger early-to-late change in correctness than users receiving random recommendations. Together, these results show how warm-start transfer and inventory-aware recommendations can support personalization for short-lived, repeatedly cold-starting cohorts.

We extend a lattice-QCD determination of the response of the quark-gluon plasma to an external magnetic field using a Taylor expansion of the pressure at zero field. A zero-net-flux “half-and-half” magnetic-field configuration permits continuous differentiation with respect to the field amplitude on gauge ensembles generated at $B=0$. The electromagnetic vector potential is chosen to minimize stochastic noise in the Taylor operators. Previous results through $O(B^2)$ are extended to $O(B^4)$ using $2+1$-flavor HISQ/tree ensembles with $m_l/m_s=0.05$ and $N_t=8$. The renormalized quadratic coefficient confirms paramagnetic behavior above the QCD crossover. The fourth-order coefficient $C_4$ is small at high temperature and is enhanced near the crossover. At $T=167$ MeV the $C_4$ result at $48^3\times 8$ lies about two standard deviations above the corresponding $32^3\times8$ one, indicating likely finite- volume effects. For $eB = 0.2$ GeV$^2$, characteristic of magnetic fields relevant to heavy-ion phenomenology, the fourth-order contribution is small compared with the leading magnetic correction at moderate and high temperatures, but near the crossover the large-volume result is about half of the central value of the quadratic term, after the vacuum-pressure subtraction. We also place the method in the context of other lattice determinations of magnetic susceptibility and clarify the relevance of finite-volume effects for the thermal subtraction. This work was completed in 2014 and preliminary results were reported at Lattice 2014.

Speech self-supervised learning aims to learn general-purpose representations for downstream speech tasks. However, current approaches rely on complex, carefully designed prediction targets. We challenge this necessity with GLaS-JEPA, a framework that directly predicts the current encoder's continuous representations at masked positions, without contrastive learning, discrete targets, or separate EMA target encoders. We prevent representation collapse using SIGReg representation-space regularization, eliminating the need for engineered target-generation mechanisms. Pretrained on 960 hours of LibriSpeech, our 57M-parameter model achieves a 6.89% WER on frozen-encoder SUPERB ASR and a 25.87% CER on slot filling, outperforming the best non-distilled sub-90M baselines by 43.1% and 22.0%, respectively. These results demonstrate that highly competitive speech representations can emerge from a radically simplified training recipe.

This article aims to raise awareness within the ESO astronomical community of the scientific opportunities offered by the upcoming Cherenkov Telescope Array, CTAO, highlighting cases where combining millimetre/submillimetre capabilities of the Atacama Large Millimeter / submillimeter Array, ALMA with the CTAO's very high energy, TeV, observations enables new avenues for discovery. Many astrophysical sources and environments exhibit emission processes detectable in both regimes; however, the synergies extend beyond shared emission mechanisms and include broader physical processes and observational connections linking millimetre, submillimetre and TeV energies.

A majority coloring of a graph $G = (V,E)$ is a coloring of $V(G)$ such that, for each vertex $v$, the number of neighbors of $v$ with the same color as $v$ is at most $deg(v)/2$. A strong majority coloring is a coloring of $V(G)$ such that, for each vertex $v$, every monochromatic subset of $N(v)$ has size at most $deg(v)/2$. The (strong) majority coloring game is a two-player Maker-Breaker-type game, in which two players Alice and Bob color the vertices of a graph $G$ alternately, maintaining the (strong) majority condition. The least number of colors such that Alice has a winning strategy in such a game is called the (strong) majority game chromatic number of the graph $G$, denoted $\mu_g(G)$ (or $\mathrm{Maj}_g(G)$ for the strong version). For the majority coloring game, we prove that $\mu_g(G) \le 3$ under the following cases: $G$ is a $2$-caterpillar, $G$ is a rooted tree with all leaves at depth $k \le 4$, and $G$ is a subdivision of some graph. The latter resolves a problem posed by Bosek–Grytczuk–Jakóbczak in 2019, who also asked whether $\mu_g(T) \le 3$ for every tree $T$. For the latter question, we discuss various difficulties that arise when natural strategies are attempted by Alice to win the majority coloring game on trees. We include a comparison with the marking game and relaxed coloring game on trees, and with the majority coloring game on locally finite acyclic graphs $G$ with $\delta(G) > 1$. For the strong majority coloring game, we compute $\mathrm{Maj}_g(C_n)$ exactly for each cycle $C_n$, $n \ge 3$. We also initiate the study of the computational complexity of the strong majority coloring game; specifically, we prove that the decision version of the Strong Majority Game Chromatic Number problem is PSPACE-complete. We also show that the Strong Majority 2-Coloring problem is NP-complete on Eulerian graphs.

We compute the scaling dimensions of monopole operators in QED$_3$ with $N$ 2-component complex fermions and Chern-Simons level $k$ to subleading order in the limit where $k,N$ are large and $k/N$ is fixed, which generalizes previous results for scalar QED$_3$ (sQED$_3$) coupled to $N$ complex scalars. In both cases, at leading large $N$ many monopole operators are degenerate, and we compute their energy splitting. We use these results to check the duality between QED$_3$ with $N=1,k=-3/2$ and sQED$_3$ with $N=1,k=2$, whose $U(1)$ monopole symmetry is enhanced to $SO(3)$. We find the scaling dimensions in each dual theory match to high precision for all charges, the lowest charge spin 1 monopoles are close to 2 as expected for the emergent $SO(3)$ current, and the degenerate second lowest charge monopoles match a prediction from a fuzzy sphere calculation, including the energy splitting. We also find similar evidence for dualities between QED$_3$ with $N=1,k=-(2m+1)/2$ and one scalar coupled to two $U(1)$ gauge fields, which has an effective description as sQED$_3$ with $N=1,k=\frac{m+1}{m}$ and $q_s=mq_f$ for $m>1$. Finally, we also find similar evidence for Son's duality between QED$_3$ with $N=1,k=0$ and a free fermion, and clarify a previously discussed match between sQED$_3$ with $N=k=1$ dual to a free fermion.

We find explicit formulas for the EHZ and cylindrical capacities of rotated $L_{p}$-ellipsoids. In addition, we provide an algorithm to estimate the EHZ capacity.

World-Action Models (WAMs) couple visual dynamics with action prediction, bringing the rich priors of pretrained video models to robotic manipulation. However, their multi-view interfaces typically tile images or concatenate tokens, leaving the geometric relationships among synchronized cameras implicit. This makes it harder to connect global scene context with the local geometry required for interaction. We introduce the Multi-View Geometry-Aware World-Action Model (MVG-WAM), which organizes these observations as related projections of one physical world rather than separate images on a canvas. Our model combines an epipolar-constrained global state with view-indexed geometric states jointly inferred from synchronized observations. Camera-aware routing supplies each video region with its corresponding geometric context and the shared global state, explicitly structuring the representation used for action prediction. We further ground the geometry-aware representation in metric scale through multi-horizon future-depth supervision, without requiring depth decoding during action rollout. MVG-WAM achieves average success rates of 99.1% on LIBERO and 92.07% on RoboTwin 2.0, demonstrating competitive performance across both benchmarks. Real-world experiments on Cobot Magic further demonstrate a 91.3% success rate across 150 trials spanning three manipulation tasks.

High-entropy stabilization chemistry is redefining materials design by transforming configurational disorder, arising from the deliberate incorporation of multiple principal cations, into a thermodynamic advantage that promotes phase stability and enables emergent functionalities. In this work, we investigate the evolution of magnetic ordering in spinel-type high entropy oxides by systematically varying the cation composition of the B site within a fixed high-entropy A-site matrix, (Ni$_{0.2}$Mg$_{0.2}$Co$_{0.2}$Cu$_{0.2}$Zn$_{0.2}$)B$_2$O$_4$. Upon introducing multicomponent B-site configurations, we uncover a strikingly linear dependence of the magnetic transition temperature (T$_C$) on the T$_C$s of the corresponding single B-site high-entropy systems. Remarkably, this trend persists even in highly complex (Ni$_{0.2}$Mg$_{0.2}$Co$_{0.2}$Cu$_{0.2}$Zn$_{0.2}$)(Cr$_{0.2}$Mn$_{0.2}$Fe$_{0.2}$Ga$_{0.2}$X$_{0.2}$)$_2$O$_4$, X = Al and Ti. Despite the material's extremely high degree of disorder, absence of a dominant magnetic ion or a straightforward superexchange pathway, detailed magnetization measurements, low-temperature X-ray magnetic circular dichroism, and neutron powder diffraction studies reveal robust long-range ferrimagnetic ordering. These results reveal an emergent predictability in ferrimagnetic high-entropy spinel oxides, where, despite extreme configurational disorder and competing interactions, robust ferrimagnetic order can arise from, rather than be hindered by, extreme configurational disorder. This establishes a pathway for predictively tuning magnetic transition temperatures in high-entropy oxides beyond conventional ordered systems.

Every day, people continuously infer situational context and adjust the way they understand and remember the world. Context, signaled by the prefrontal cortex, is known to modulate working memory and episodic memory, but the algorithmic understanding of this modulation remains limited. Here, we train a recurrent neural network (RNN), augmented with an episodic memory buffer, to infer context using Bayesian inference as it continuously makes predictions of upcoming scenes while watching naturalistic movies. When the inferred context modulates the RNN's recurrent connectivity (the basis of working memory) in a low-rank manner, the model's activity patterns best match neural responses in human participants who watched the same movies during fMRI. Context also modulates episodic memory retrieval, such that the model retrieves memories based on not only content similarity but also context similarity. This is implemented as a key-value system with self-attention, designed to additionally encode context and retrieve context-congruent memories. The resulting model not only better resembles human brain representations but also learns to retrieve memories like humans much faster than a model without context modulation. Together, our findings suggest a computational mechanism by which context modulates information maintenance and long-term memory retrieval in naturalistic environments.

We prove that if $f_1,\ldots,f_N$ is an uniformly bounded orthonormal system, then for all $p\in[1,2)$ there is an estimate for its Kolmogorov widths: $d_n(\{f_1,\dots,f_N\},L_p) \gtrsim \min\{1,n^{-1/p}N^{1/2}\}$, $n\le N/2$. In the regime $n\asymp N$ this provides the lower bound $N^{-\alpha_p}$ with a sharp exponent $\alpha_p:=1/p-1/2$. Besides that, it follows that a good approximation of such ONS requires the dimension $n\gtrsim N^{p/2}$. Our second result is an approximation theorem for Fourier matrices of finite abelian groups (this includes the usual DFT matrices). For every $\eta>0$ there is $a=a(\eta)>0$ such that any Fourier matrix $F$ of sufficiently large order has an approximation of rank $N^{1-a}$ with the row-wise $\ell_1$-error at most $N^{1/2+\eta}$; the exponent $1/2$ is sharp. Consequences include the approximation of the same kind for circulant matrices; the approximation of the trigonometric functions $\exp(2\pi i\langle \lambda,x\rangle)$, $\lambda\in \Lambda=K\cap\mathbb{Z}^d$, $K$ is symmetric convex, with dimension $|\Lambda|^{1-a}$ and an optimal error $|\Lambda|^{-\alpha_p+\eta}$; bounds for widths of weighted Wiener classes.

Self-supervised learning (SSL) by predicting in latent space, without generating the input data itself, learns highly abstract, useful representations. Intuitively, this success is often attributed to its ability to discard nuisance information that is irrelevant to prediction. However, this poses a conundrum: both stochastic variation in a prediction-relevant latent signal and true nuisance make observations partly unpredictable; how could they be distinguished? Surprisingly, we prove that common SSL methods can achieve exactly this, by implicitly instantiating a latent-variable model with stochastic dynamics and observation-private nuisance. We trace their ability to recover the stochastic signal to two complementary principles: Predictive mutual information maximization ensures that representations retain the information needed for prediction, while latent distribution matching constrains how this information is encoded, thereby making the retained signal identifiable. We confirm this identifiability result in simulations for Gaussian predictors, which recover the true signal up to an affine transformation even in dynamic, nuisance-laden environments.