Papers

Speech understanding in noise remains challenging for hearing-aid users, particularly in the presence of competing speakers. Conventional hearing aids typically perform speech enhancement (SE) and hearing-loss compensation in separate stages, which may cause enhancement errors and signal distortions to carry over to the amplification stage. Moreover, audio-only SE often provides limited benefits under competing-speech conditions because the target and interfering speech share similar acoustic characteristics, making them difficult to separate. To address these limitations, we propose AV-NeuroAMP, an end-to-end audiovisual framework that integrates noisy speech, target-talker video, and the listener's audiogram to jointly perform SE, personalized amplification, and dynamic-range compression. We further introduce audiogram-conditioned feature-wise linear modulation (AC-FiLM) to effectively incorporate listener-specific hearing profiles. Objective evaluations showed that AV-NeuroAMP outperformed conventional amplification, the audio-only NeuroAMP model, and two-stage systems on an in-domain English test set, and that these improvements were retained on an unseen Mandarin test set. Listening tests involving normal-hearing participants under simulated hearing loss and listeners with hearing loss further demonstrated improvements in speech quality and intelligibility, with the greatest benefits observed under competing-speech conditions. These findings support end-to-end audiovisual personalized amplification as a promising approach for improving hearing-aid performance in challenging acoustic environments.

Transformers provide a state-of-the-art modeling framework, yet poor calibration limits their reliability in safety-critical applications. A promising direction addresses this issue by interpreting attention as a Gaussian process (GP) posterior, which enables principled uncertainty calibration but incurs cubic complexity in sequence length due to the inversion of the kernel; although decoupled GP variants reduced the cost to quadratic, the computation remains prohibitive in practice. In this paper, we propose the plug-and-play random Fourier feature Gaussian process attention (RFF-GPA) module, which represents the attention as a GP with a stationary kernel approximated by random Fourier features. This low-rank approximation results in linear-time complexity for approximating the posterior mean and variance, making it far more scalable compared to previous work. Empirical results on multiple real-world datasets show that our attention module improves calibration while maintaining predictive accuracy, and simultaneously reduces computational complexity to linear in the sequence length.

While unlearning seeks to negate undesired capabilities acquired through learning, little research has examined how the way models learn shapes their subsequent unlearning. In this paper, we investigate this connection from the perspectives of memorization and generalization, the two most representative yet competing strategies that models employ during training. We first classify memorization- and generalization-heavy models using grokking in modular addition and compare their responses to unlearning, showing that the latter suffer greater retain damage, i.e., a larger performance drop on the retain set. Furthermore, we conduct a finer-grained analysis by introducing bucketed modular addition, in which the respective contributions of the two strategies can be explicitly controlled across the memorization-generalization spectrum. In this setup, we reaffirm that the same trend persists and is nearly monotonic. We further demonstrate that this relationship also holds in LLM unlearning across verbatim and factual recall settings. Finally, we provide two practical insights for developing better unlearning methods, highlighting the importance of accounting for learning dynamics in unlearning.

We address the Subset selection problem for matrices, where the goal is to select a subset $\mathcal{S}$ of $k$ column indices from a \enquote{short-and-fat} matrix $X \in \mathbb{R}^{m \times n}$, such that the sampled submatrix $X_{\mathcal{S}}$ has $\|X_{\mathcal{S}}^\dag X\|_F$ as small as possible. Our approach is centered on volume sampling, which attains the tightest known bound on this objective in expectation. As our primary contribution, we propose a new deterministic algorithm, Forward derandomized volume sampling (FDVS), which provably attains this bound and has asymptotic complexity $O(nkm)$. In contrast, the complexity of all previously known algorithms with this guarantee scales at least quadratically in $n$ when $k \ll n$, making FDVS particularly attractive for very wide matrices. In addition, we systematically structure the landscape of volume sampling methods: we propose a simple $O(nm^2)$ algorithm for exact forward volume sampling, show that a known deterministic method is in fact a derandomization of Reverse iterative volume sampling, and derive a modification of the latter that avoids repeated SVD downdating, along with a fast greedy forward variant. The algorithms are verified and compared in numerical experiments involving optimal experimental design, sensor placement, and DLRA-DEIM.

We propose a finite-temperature Gaussian-fluctuation formalism for response functions within ghost rotationally invariant slave-boson (gRISB) theory and apply it to the two-dimensional Hubbard model. The gRISB magnetic phase boundary and static spin and charge susceptibilities capture the trends of dynamical mean-field theory (DMFT), significantly improving upon standard RISB and the random phase approximation (RPA). At the RISB level, our formalism recovers the Kotliar-Ruckenstein response. The dynamical susceptibilities resolve paramagnons, low-energy sound modes, and high-energy Hubbard excitations. Because they are evaluated directly on the real-frequency axis, the results avoid Matsubara truncation, stochastic sampling, and numerical analytic continuation. We also introduce a conserving quasiparticle approximation that expresses the response in terms of dressed Green's functions and vertex corrections.

Skin cancer is a major global health concern, and early detection and accurate lesion delineation are important for effective diagnosis and treatment planning. Automated skin lesion analysis can assist dermatologists, with lesion segmentation serving as a fundamental step in computer-aided diagnostic systems. Conventional deep learning-based segmentation models typically rely on centralized training, where images and their corresponding segmentation masks are collected on a central server. Such data aggregation raises privacy concerns in medical applications and requires substantial centralized computational resources. To address these limitations, we investigate the feasibility of federated learning for privacy-preserving skin lesion segmentation. The training and validation sets of the ISIC 2018 Skin Lesion Segmentation Challenge dataset are used to simulate a distributed learning environment and develop a federated segmentation model. The resulting model is evaluated on the ISIC 2018 test set and the PH2 dataset to assess its performance and generalizability. Experimental results demonstrate that the federated model achieves performance comparable to centralized training while consistently improving upon the locally trained models. These findings demonstrate the potential of federated learning for collaborative skin lesion segmentation without requiring centralized aggregation of medical images.

Cooperative perception improves autonomous driving by sharing complementary observations among vehicles and roadside infrastructure for 3D object detection. However, practical deployment is constrained by limited bandwidth and unreliable cooperation, where packet loss, transmission delay, and spatial misalignment jointly degrade the cooperative feature stream. Existing methods often reduce communication cost or compensate for one degradation type, leaving coupled disturbances insufficiently addressed. To address this problem, we propose Sparse2comm, a bandwidth-efficient and robust cooperative 3D object detection framework that treats unreliable cooperation as progressive restoration over degraded cooperative features. Sparse Feature Encoding first encodes communication as randomly mask-sampled foreground features transmitted by collaborating agents, from which the ego vehicle reconstructs dense semantic representations. This sparse-to-dense mechanism learns to infer missing object-centric content from sparse observations, enabling ultra-low-bandwidth communication and packet-loss recovery within the same representation. On the semantically restored features, Latency-Aware Alignment predicts motion flow to compensate delayed messages, and Self-Calibrating Fusion estimates residual spatial offsets in a self-supervised manner before adaptive cross-agent fusion. Sparse2comm therefore restores semantic completeness, temporal consistency, and spatial alignment in an ordered pipeline. Extensive experiments on DAIR-V2X, OpenV2V, and V2V4Real show that Sparse2comm maintains competitive clean accuracy and consistently improves robustness under individual and mixed real-world degradations. Compared with the selective feature communication baseline Where2comm, Sparse2comm improves mixed-setting AP@0.5/AP@0.7 by +20.15/+11.79, +12.66/+11.07, and +15.36/+12.61 on the three datasets, respectively.

In this paper, we investigate the Takagi-van der Waerden function, $$ T_r(x) = \sum_{n=0}^{\infty} \frac{\phi(r^n x)}{r^n} ,\quad x\in [0,1], \quad r \geq 2, $$ where $\phi(x)={\rm dist}(x,\mathbb{Z})$ represents the distance from $x$ to the nearest integer. Lagarias and Maddock[13] introduced the notion of local level sets for the classical Takagi function $T_2$. They proved that if $y$ is uniformly distributed over the range of $T_2$, then the expected number of local level sets contained in the level set at height $y$ equals $3/2$. We define the corresponding local level sets for every integer $r\geq2$ and prove that, if $y$ is uniformly distributed over the range of $T_r$, then the expected number of local level sets contained in the level set at height $y$ equals $(r+1)/r$ if $r$ is even, and $(r+1)/(r-1)$ if $r$ is odd.

Retrieval-Augmented Generation (RAG) systems are vulnerable to prompt-injection attacks embedded in retrieved content. We introduce RAG-PIBench, a benchmark for RAG-style prompt-injection detection containing 4,876 contextual examples across frozen train, validation, and protected-test splits. Using a leakage-aware construction pipeline and strict evaluation protocol, we compare keyword-based, semantic-reference, TF-IDF, and transformer-based detectors. DistilBERT achieves the best protected-test performance (F1 = 0.896, PR-AUC = 0.968), while TF-IDF SVM and logistic regression remain competitive. Our results demonstrate the value of leakage-aware benchmark design and strong sparse baselines for reliable prompt-injection detection in RAG systems.

Handcrafted preprocessing is widely employed in automated dermoscopic analysis to suppress imaging artifacts and enhance lesion visibility. Nevertheless, its actual contribution to modern real-time models remains unclear, particularly when evaluation protocols do not adequately control correlations among images of the same lesion. This study presents a leakage-controlled, lesion-disjoint evaluation of dermoscopic preprocessing and augmentation for joint multi-class lesion classification and instance segmentation using a fixed nano-scale YOLO26 segmentation model (YOLO26n-seg). From HAM10000 (10,015 images), quality control yields 10,013 valid image-mask pairs from 7,468 unique lesions, partitioned into mutually exclusive sets by lesion identity. With the architecture, resolution, training budget, and evaluation protocol held fixed, we compare minimally processed images plus online augmentation against offline class balancing, DullRazor-CLAHE preprocessing, and raw-processed hybrid views, over three random seeds. On the lesion-disjoint test set, the raw baseline achieves a mask mAP$_{50:95}$ of $0.5636 \pm 0.0234$, a Dice score of $0.9356 \pm 0.0024$, and a macro-F1 score of $0.6917 \pm 0.0202$. Offline augmentation does not improve the mean performance, while the combined and hybrid strategies reduce both class-aware segmentation and classification accuracy. At only 2.69 million parameters, the model runs at approximately 50 frames per second. Under a leakage-controlled, lesion-disjoint protocol with all non-input factors held fixed, minimally processed dermoscopic images combined with standard online augmentation deliver a better accuracy-efficiency trade-off than increasingly complex deterministic preprocessing, which yields no consistent joint benefit across three seeds on HAM10000.

For an integer $g$ not equal to $-1$ or a square, we study the least prime $p_g$ such that $g$ is a primitive root modulo $p_g$. We show that, assuming the Generalised Riemann Hypothesis, all such $g$ with $|g|\in[N,2N]$ have $p_g\leq(\log |g|)^{3.44}$ with at most $O((\log N)^{2.44})$ exceptions. This can be directly compared to the uniform bound $p_g\leq(\log |g|)^{19}$ proven recently by Fan and Pollack. To obtain our result, we prove a new variant of Gallagher's larger sieve, which may be of independent interest. In addition to our conditional result, we discuss other “almost all" bounds for $p_g$ that hold unconditionally.

We characterize when a banded totally positive matrix admits a positive bidiagonal factorization. In finite dimension such a factorization always exists. For semi-infinite matrices of arbitrary finite bandwidth, existence is equivalent to convergence of a finite family of positive series determined by the Metelmann factors. Their reciprocals are the limiting values of scalar Jacobi continued fractions. We also prove that every bounded banded totally positive matrix can be approximated in operator norm by matrices admitting positive bidiagonal factorizations, through perturbations confined to a fixed initial block. For banded Toeplitz matrices, the negative roots of the polynomial symbol give a global factorization with bounded factors. Ordered partitions of these roots determine explicit mixed-type multiple orthogonal polynomials. Applying the existing spectral Favard theorem, we compute their matrix of positive measures: its support is one interval, and the measure is purely absolutely continuous with an algebraic density of rank one. These formulas also give transition and first-return laws for the associated killed random walks.

A $d$-simplex-cluster is a collection of $d+1$ distinct $k$-element sets with empty total intersection, nonempty intersection for every $d$ members, and union of size at most $2k$. We prove the simplex-cluster conjecture of Keevash and Mubayi, a common strengthening of the Erdős–Chvátal simplex conjecture and Mubayi's cluster conjecture. More precisely, for integers $k>d\ge2$ and $n\ge k(d+1)/d$, every family of $k$-element subsets of an $n$-element set containing no $d$-simplex-cluster has at most $\binom{n-1}{k-1}$ members. Equality holds if and only if the family consists of all $k$-element subsets containing a fixed point.

We construct Yang-Mills theories with projected curvature kinetic terms, developing their analogy with MacDowell-Mansouri gravity and their connection to nonlinear sigma models. We begin with models exhibiting explicit symmetry breaking and use them to motivate gauge-covariant Higgs extensions. For $SO(2n)$, an orthogonal complex structure defines a degenerate kinetic operator with image $\mathfrak{su}(n)$, while the adjoint Higgs vacuum has stabilizer $U(n)$. We compare the fixed-projector theory with gauge-covariant formulations involving either unrestricted adjoint Higgs fields or Higgs fields constrained exactly to the orbit $SO(2n)/U(n)$. We express the Stelle-West Higgs-gravity action through a single Clifford involution whose complementary projectors select the Lorentz curvature, isolate the radial Higgs derivative, and construct the induced spacetime metric. This formulation makes the common projector structure explicit and clarifies the distinct roles of curvature projection, the internal dual, and the coframe. In the orbit-constrained Yang-Mills theory with a positive Higgs kinetic coefficient the coset connection is auxiliary and the central connection drops out of the action. A nonlinear Dirac analysis establishes a regular weak-curvature branch with $2(n^2-1)$ physical degrees of freedom in four dimensions, locally equivalent to $SU(n)$ Yang-Mills theory. In the unrestricted theory, normal Higgs displacements can change the kinetic rank, so this degree-of-freedom count does not automatically extend beyond the constrained field space. We also characterize possible nontrivial algebraic branches and introduce projected Pontryagin theories and a full-rank equivariant completion.

This article is a geometric rediscovery of the singular value decomposition, with a further claim: the construction it builds is the machinery behind much of machine learning. The same argument that answers an idle question about ellipses is the algorithm behind principal component analysis, kernel methods, and PageRank, and it is not only the results that transfer but the proofs themselves, run as procedures. The usual introduction states $A = U\Sigma V^T$ and justifies it via the spectral theorem applied to $A^T A$. This is correct but unilluminating, since it assumes a powerful theorem to reach a result that is, in the end, about ellipses. Part I reverses the order. A linear map sends the unit circle to an ellipse; one asks which input directions map to its axes, and finds, example after example, that they are perpendicular. In the plane this can be watched: rotate a frame, track how far its images are from perpendicular, and a sign change forces a frame where they are exactly perpendicular, which is also where the map stretches hardest. Maximizing the stretch and recursing generalizes this to n dimensions, with singular values falling out in order, and the construction proves the spectral theorem rather than assuming it. Part II puts each construction to work: maximize-and-recurse becomes the power method and PageRank; the lemma locating the maximizer becomes the stopping rule of gradient descent; the duality between $A^T A$ and $A A^T$ becomes the transport at the heart of kernel PCA. Each connection is stated with its boundary, saying what the decomposition supplies and where another idea takes over. Prerequisites are the standard sophomore sequence, and the worked examples are small enough to check by hand.

Tabular foundation models (TFMs) can classify the nodes of a graph without training on it, by reading node and neighborhood features as table rows next to labeled context rows. Work in this line reports predictive performance, not conformal coverage or prediction-set size. To our knowledge, we give the first reliability study of the setting, with TabICL as the TFM and half of each graph as labeled context. As for any predictor fixed before calibration, a frozen in-context predictor makes split conformal prediction exactly valid in finite samples, with no training, validation fold, or tuning on the target graph. An audit across ten graphs then shows that the training-free TabICL posterior has lower expected calibration error (ECE) than GCN with temperature scaling (GCN+TS) on nine of them. Its mean ECE over the ten graphs is 0.019, about 35 percent below the 0.029 of GCN+TS. We also introduce HG-DAPS, a training-free diffusion score whose homophily gate reads only the in-context labels, so the guarantee still holds. Relative to adaptive prediction sets (APS), it reduces mean set size by 5.8 to 17.1 percent on six homophilous graphs and changes it by under 1 percent on four heterophilous ones. On two binary, class-imbalanced graphs, a pre-registered trap case shows that gating on raw rather than adjusted homophily lowers coverage among low-homophily nodes by 0.27 and 0.12. Marginal coverage stays at the nominal 0.90 and masks this drop.

Sequence-level power sampling has recently emerged as a training-free approach to reasoning by sampling from a sharpened output distribution of a base large language model (LLM). Nevertheless, existing methods typically sharpen the base model distribution uniformly across queries, overlooking variations in query difficulty and in how well the base model already handles each query. The goal of this work is to equip power sampling with query adaptivity. Theoretically, we show that the benefits of further sharpening are determined by the self-reward gap between correct and incorrect responses. Based on this insight, we propose Adaptive Power Sampling (APS), which adjusts the sharpening exponent on a per-query basis at test time using the relationship between answer agreement and the model's self-reward. Experiments across diverse reasoning tasks, including MATH500, HumanEval, and GPQA, show that APS consistently outperforms power sampling with a fixed sharpening exponent, without additional training.

The Gram-Schmidt Walk is a randomized vector-balancing algorithm whose subgaussian guarantees support applications in discrepancy, experimental design, and data compression; however, these theoretical guarantees are established in exact arithmetic, whereas implementations must approximate least-squares directions, boundary updates, and sampling probabilities in finite precision. This is important because small numerical errors can change which coordinates freeze and thereby alter the subsequent trajectory. We analyze the concentration of the perturbed Gram-Schmidt walk directly under bounded, potentially biased and history-dependent errors. For $n$ input vectors of Euclidean norm at most one, we obtain a modified MGF bound depending on key error sources which recovers the original result as the error goes to zero. We also construct a full-column-rank instance in which bounded update errors produce bias of order $\min\{n^2\varepsilon,n\}$, showing that updates accumulate error unavoidably under this model. Finally, we validate our findings in a variety of settings by ablating on the bit precision and problem size.

Reinforcement learning (RL) has become a standard tool for post-training language models on reasoning tasks, where the policy is updated by reward feedback while exploring the space of responses. Despite its empirical success, theoretical understanding of RL post-training remains limited, in particular of why on-policy exploration combined with a neural reward model is effective. In this paper, we address this question by modeling the reward as a hierarchical function on the response space: the reward consists of infinitely many local components, each of which becomes relevant only after the preceding ones have been resolved. We show that a natural Transformer-based actor–critic algorithm, which alternates between sampling from the current KL-regularized policy, fitting a Transformer critic to the observed rewards, and updating the policy, achieves the minimax optimal rates in the query budget and in the regularization strength up to logarithmic factors, and is minimax optimal for a fixed number of prompts. In contrast, we prove that sampling from the fixed reference distribution, as in offline reward modeling, can limit regret decay to a logarithmic rate. These results show that on-policy exploration progressively zooms in on the region where the reward is concentrated, and quantify its benefit for RL post-training.

Streaming video-language models must decide not only what to answer, but whether the evidence needed for the current question has arrived. Existing systems learn that decision as a separate trigger; we ask whether an unmodified model already computes it. We show that frozen VideoLLMs carry a linearly readable evidence-readiness signal, labelled from timestamped evidence rather than from model output. It decodes in all seven models of a shared byte-identical evaluation (AUROC 0.733-0.905 under the strictest not-ready sampling, where a fitted clock is near chance), and a probe fitted without any of a benchmark family's footage still reads that family. It is question-conditioned: on byte-identical windows, changing only the question reverses the readout on 66.1% of pairs, while every question-blind control is at chance by construction. The model can answer incorrectly and still encode readiness: AUROC remains 0.722 among wrong answers. Readiness also beats uncertainty estimators and their supervised combination on latency-matched answer selection, and tracks independent human judgments more closely than confidence. Released streaming triggers are also linear readouts, yet a trained trigger read on its own base model's activations is approximately orthogonal to readiness and decodes it far less accurately than a probe. We turn the readout into Readiness Gating, an answer-timing policy that improves accuracy by up to +9.75 pp at matched video duration with negligible computational overhead. How much it gains varies with the accuracy headroom the task makes available: across 26 configurations the gain tracks that headroom, and an intervention that moves it over identical pixels moves the gain with it.

Activation steering has gained popularity as a lightweight inference-time debiasing technique for large language models. However, prior work reports that steering vectors generalise poorly, with unintended effects on model performance and limited transfer to new datasets. Our work analyses what the debiasing direction used for activation steering actually encodes, in order to shed light on its inconsistent performance. We study the linear debiasing direction obtained by contrasting the activations of anti-biased and biased prompts, and evaluate it as a steering intervention across bias and general knowledge benchmarks. We find that this direction is dominated by model confidence, pointing from regions of high to low-probability tokens in activation space rather than encoding a meaningful representation of model bias. Steering along it does reduce measured bias, but this is a consequence of reducing model confidence: on QA benchmarks we find that this steering drives the model to abstain from answering, with a side effect of improving fairness metrics. Our experiments show that model confidence is the dominant separating factor between biased and anti-biased prompts in hidden space, indicating that isolating a linear representation of bias which is disentangled from model confidence is difficult and steering-based debiasing results should be interpreted with care. In short, steering appears to reduce bias, not by correcting the model's underlying preferences, but by making it less confident, even on tasks unrelated to bias.

Businesses regularly offer gift card programs to drive customer spending and increase engagement. A central question is how much incremental revenue these programs generate, and which channels drive it most efficiently. Measuring the incremental revenue associated with a gift card program is a challenging problem in causal inference, requiring a firm to infer how much each customer would have spent if they never received a gift card. Observational data on past customer purchasing behavior reveal possession of a gift card only when a customer makes a purchase, thus leaving a customer's treatment status systematically censored. In this paper, we develop a novel data fusion approach to overcome this missing data challenge. We identify and estimate incrementality by combining a large observational dataset with a smaller experimental dataset from a different population. Our approach relies on a mild transferability condition, which posits that the conditional relative treatment effect of gift card receipt on the decision to purchase is invariant across the two populations. We develop a flexible, machine learning-based estimator for the incremental revenue and establish its asymptotic normality. We apply our estimator across both first- and third-party channels through which Airbnb distributes gift cards, finding heterogeneity in incrementality across segments of the population. In particular, we find not only that third-party channels are more incremental than first-party ones, but also that "self-gifters" (i.e., customers likely to have purchased their own gift cards) are more incremental than the broader population.

An Andreev spin qubit stores a single unpaired spin, yet its coupling to the superconducting phase can be dominated by coherent virtual quasiparticle pairs. We develop a continuum variational description of the odd-parity doublet in a quantum-dot Josephson junction with Coulomb interaction and spin-dependent background tunneling. Retaining configurations with up to two Bogoliubov quasiparticles provides compact analytical expressions for energies and wavefunctions. We benchmark the results against numerical renormalization group calculations and identify sources of error. We obtain closed-form expressions for the conventional and spin-dependent Josephson couplings. Despite their small probability, two-quasiparticle configurations, through coherence with the zero-quasiparticle component, supply exactly half of the leading spin-dependent Josephson coupling even without Coulomb repulsion and dominate it as the repulsion grows. Spin-orbit-induced spin transfer between the dot and the leads produces a Josephson-current contribution that is common to both qubit states and finite at zero phase bias when the dot and lead Zeeman energies differ. We also map the localization of the virtual quasiparticles and the spin density in the leads. The results connect strongly correlated quantum-dot models to effective Hamiltonians used for Andreev-spin-qubit circuits and clarify how the wavefunction structure determines their couplings.

Let $M$ be a closed connected Riemannian manifold and let $\mathcal P_s$ be the fractional perimeter on $M\times\mathbb R^k$ defined through its heat semigroup, where $k\ge1$ and $0<s<1$. We prove that the straight cylinder $M\times B_R$, with $B_R\subset\mathbb R^k$ a ball of radius $R$, uniquely minimizes $\mathcal P_s$ among all measurable sets of the same volume, up to translations in $\mathbb R^k$ and null sets, whenever $R\sqrt{\lambda_1(M)}$ is at least an explicit threshold depending only on $k$ and $s$. Here $\lambda_1(M)$ is the first positive eigenvalue of the nonnegative Laplace–Beltrami operator $-\Delta_M$ on $M$. Under a larger threshold, the perimeter deficit controls both the deviation of the slice volumes from their mean and the square of the asymmetry relative to straight cylinders. For flat tori this gives large-volume uniqueness of straight cylinders for the periodic fractional perimeter, including singly periodic sets in every ambient dimension. The proof combines spectral reduction on $M$ with rearrangement of the Euclidean slices and a one-variable comparison between an interaction gain and a concavity loss.

Network pruning can reduce parameter redundancy in robotic policies. However, generic pruning criteria are tailored for image recognition tasks and commonly designed to preserve weight magnitude, local reconstruction, or language-model likelihood rather than closed-loop action behavior. Directly applying these pruning algorithms to robotic tasks yields unsatisfactory performance. In this paper, we propose Loss-Conditioned Activation-Moment (LCAM) pruning, a training-free method for unstructured pruning of pre-trained robotic manipulation policies. Specifically, we first rank connections using row-normalized weight contribution, activation moments measured on calibration demonstrations, and the sensitivity of output directions to the action-prediction loss. We further design a self-recursive coarse-to-fine procedure: importance is recalibrated after each nested coarse pruning stage, while held-out offline action distortion guides fine-grained budget allocation after a sparsity knee. Our algorithm is free from costly recovery training and simulator rollouts after pruning. Experiments on three LIBERO suites with competitive robotic policies, together with evaluations on OpenVLA, show that LCAM attains competitive performance across a broad range of pruning ratios. Notably, on LIBERO-Object with OpenVLA, our LCAM achieves 84.0% success at 50% unstructured pruning, retaining over 90% of the dense policy's success rate. Promising results on real-world robotic ping pong further demonstrate the effectiveness of our pruning algorithm.

While HCI increasingly examines AI-safety for youth, the literature lacks a comprehensive view of what risks have been identified, how they are addressed, and whether proposed protections work in-practice. We systematically reviewed 100 empirical HCI studies involving children and youth interacting with or exposed to AI across schools, homes, care settings, and public services. Using the YAIR taxonomy for risks and the MIT Mitigation Taxonomy for countermeasures, we map which risks have been identified, whether each risk is addressed by countermeasure(s), and whether each countermeasure for that risk is implemented and even evaluated. The risk-countermeasure mapping shows that most risks are matched only with proposed/ideated countermeasures; few countermeasures have been implemented, and fewer still evaluated; and existing evaluations often measure technical performance rather than protection from harm. We identify where coverage is absent, where safeguards remain untested, and propose concrete directions for HCI research to strengthen youth AI-safety.

Sequential test-time training adapts a memory network through successive updates, each computing an inner-loop gradient based on the network's previous state. Intuitively, this state dependence should allow each update to account for what the memory has already learned and better incorporate new information. However, we find that this expected advantage does not consistently materialize in nonlinear memories: a fixed-base parallel TTT baseline outperforms its serial counterpart. Our exploratory experiments point to a key underlying difficulty: nonlinear memories can be harder to optimize than linear ones within a single pass over the sequence. To alleviate this optimization difficulty, we introduce DeltaTTT, which replaces joint inner-loop optimization of a two-layer memory network with layerwise learning. Each layer is assigned a local prediction target and updated through a state-dependent delta rule. This formulation retains a nonlinear readout while enabling chunkwise parallel computation. Experiments on DeltaNet and LaCT backbones show improvements in language modeling and retrieval over their recurrent baselines.

Concept bottleneck models (CBMs) make predictions inspectable and intervenable by routing them through human-interpretable concepts, but originally required concept annotations. Annotation-free variants remove this requirement, but typically use large concept vocabularies, static at both training and inference, producing bottlenecks larger than any task or prediction needs and harder to inspect. We propose AnyBottle, a single recipe for building compact, task-specific CBMs. AnyBottle assumes only a frozen backbone and an unsupervised concept pool, such as a sparse autoencoder. A black-box teacher trained on the same backbone then guides selection: each round adds the concept that best explains the bottleneck's current failures, with candidates restricted to regions of teacher/student disagreement. Trained with nested dropout over this selection order, the final bottleneck predicts accurately from any concept prefix, so inference spends fewer concepts on inputs it is confident about early and more on hard ones. Since no stage is modality-specific, a new domain and task requires swapping only the backbone and concept pool. Across six vision and two text datasets and two teacher paradigms, AnyBottle yields bottlenecks with fewer concepts and higher concept consistency than annotation-free baselines, while staying close to the black-box reference. Overall, AnyBottle shows that going annotation-free need not mean going large: a small, discovered vocabulary can be as expressive as a much larger, fixed one.

Orbitronics has emerged as a promising approach for generating orbital currents using low cost, abundant materials through the orbital Hall effect. Since orbital current cannot directly interact with the local magnetization of a conventional ferromagnet, a non-magnetic layer with strong spin orbit coupling is required for converting orbital current into spin current. However, the propagation of orbital current across the interface between an orbital source and an orbital to spin converter remains poorly understood. Here, we tuned the Ru/Pt orbital interface from an abrupt interface to a graded interface by introducing a controlled RuPt intermixed layer and quantified the resulting damping like torque efficiencies. A small, intermixed thickness of 0.26 nm enhances the damping like torque efficiency by 15% compared with the abrupt Ru/Pt interface and by 70% compared with the Pt reference. Further increasing the intermixing thickness progressively suppresses the torque efficiency. We attribute this nonmonotonic behavior to the competition between enhanced orbital transmission through a gradual crystal field transition and orbital current dephasing due to alloy disorder and scattering. Our results demonstrate that the orbital-source/converter interface is an active element for controlling orbital transport and provides an experimentally accessible approach for optimizing orbital current transmission and torque generation in orbitronic heterostructures.

Reconfigurable pixel antenna arrays provide additional electromagnetic (EM) degrees of freedom for wireless transmission by enabling hardware-level adaptation of radiation responses. In practical arrays, inter-element mutual coupling (MC) alters the switch-state-dependent responses, causing MC-unaware designs to be mismatched with the realized coupled-array channels. To address this issue, we propose an MC-aware EM and signal-processing co-design framework for downlink multi-user multiple-input single-output (MU-MISO) systems. The full pixel antenna array is modeled as a unified multi-port circuit network that jointly characterizes the feed ports and switch-controlled pixel ports. By combining the impedance matrix extracted from Computer Simulation Technology (CST) Studio Suite simulations with the full-array open-circuit far-field responses, we construct a switch-state-dependent EM-domain channel that explicitly captures both intra-antenna pixel coupling and inter-element MC. Based on this model, we formulate a sum-rate maximization problem over the binary switch-state vector and the transmit precoder. To address the resulting mixed-integer non-convex problem, we develop an alternating optimization algorithm in which the binary switch-state vector is updated via sequential exhaustive block optimization (SEBO), while the transmit precoder is optimized using fractional programming (FP). CST-based simulations show that the proposed MC-aware design outperforms MC-unaware, MC-free, and conventional patch-array baselines. These results demonstrate that accurate MC modeling is essential not only for reducing model mismatch, but also for exploiting the coupled embedded responses of practical pixel antenna arrays.