coze_research_feed.xml

<?xml version='1.0' encoding='utf-8'?>
<rss xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/" version="2.0"><channel><title>Search Agenda Research Feed</title><link>https://careyourcake.github.io/my-research-feed/coze_research_feed.xml</link><description>Research papers matched to: sim-to-real, invariant representation, visual representation, invariant, feature</description><language>en-us</language><lastBuildDate>Fri, 14 Aug 2026 06:48:12 +0000</lastBuildDate><atom:link href="https://raw.githubusercontent.com/careyourcake/my-research-feed/main/coze_research_feed.xml" rel="self" type="application/rss+xml" /><item><title>Langlands Duality and Invariant Differential Operators: the Case SL(2n+1)</title><link>http://arxiv.org/abs/2608.13400</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13400</guid><pubDate>Thu, 13 Aug 2026 15:59:30 +0000</pubDate><dc:creator>V. K. Dobrev</dc:creator><description>Authors: V. K. Dobrev

Recently we started building a bridge between two cases of Langlands duality. The latter is one of the most influential topics in mathematical research. It has many different appearances and influential subtopics. Yet there is a topic that until now seems unrelated to the Langlands program. That is the topic of invariant differential operators. That is strange since both items are deeply rooted in Harish-Chandra's representation theory of semisimple Lie groups. We started with the case of the group ~$SL(2n)$. In the present paper we deal with the group $SL(2n+1)$. The two mentioned groups are similar, but their representation theories is rather different.</description><category>math.RT</category><category>math-ph</category><category>math.QA</category></item><item><title>Tracing Methamphetamine abuse in under-treatment drivers: How biomechanical and oculomotor features help detect at-risk drivers?</title><link>http://arxiv.org/abs/2608.13054</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13054</guid><pubDate>Thu, 13 Aug 2026 10:18:02 +0000</pubDate><dc:creator>Hamed Salmanzadeh, Alireza Mortezapour, Iman Tahbazzadeh Moghaddam, Farshid Ipackchi, Payam Abedinzadeh, Samira Teimoori</dc:creator><description>Authors: Hamed Salmanzadeh, Alireza Mortezapour, Iman Tahbazzadeh Moghaddam, Farshid Ipackchi, Payam Abedinzadeh, Samira Teimoori

While the detrimental impacts of driving under the influence of stimulants such as methamphetamine are well-documented, the driving performance of individuals currently under-treatment has received considerably less attention. This study compared the behavior of individuals with a history of stimulant abuse (across two distinct treatment phases) with a control group of healthy drivers using a driving simulator. Oculomotor and biomechanical data were continuously collected via an eye-tracker and a Kinect sensor, respectively. These parameters were utilized to train a K-Nearest Neighbors (KNN) classification model designed to detect high-risk behavioral patterns in drivers undergoing methamphetamine rehabilitation. Through the evaluation of various feature combinations and neighborhood configurations, the optimized model successfully discriminated between normal drivers and those with a history of abuse with an accuracy of 90%. Detecting at-risk drivers through technologies embedded in Advanced Driver Assistance Systems (ADAS) by continuously monitoring physiological and behavioral parameters, facilitates a proactive safety strategy. Issuing real-time alerts to the driver, passengers, and external monitoring networks can ultimately mitigate the risk of traffic collisions.</description><category>cs.HC</category></item><item><title>SAEVerbalizer: Generating Explanations for Sparse Autoencoder Features via Representation Verbalization</title><link>http://arxiv.org/abs/2608.13538</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13538</guid><pubDate>Thu, 13 Aug 2026 17:54:11 +0000</pubDate><dc:creator>Weihan Meng, Hongzhu Guo, Yi Jing, Dewen Liu, Zijun Yao, Xiaozhi Wang, Lei Hou, Juanzi Li</dc:creator><description>Authors: Weihan Meng, Hongzhu Guo, Yi Jing, Dewen Liu, Zijun Yao, Xiaozhi Wang, Lei Hou, Juanzi Li

Sparse autoencoders (SAEs) are proposed to extract numerous features from large language model (LLM) representations, yet explaining these features still relies primarily on external observation. This reliance leads to superficial explanations inferred from observed model behavior and computational inefficiency from collecting such behavioral evidence at scale. We introduce SAEVerbalizer, a framework that injects SAE decoder directions into an LLM's representations and fine-tunes the LLM's downstream layers to generate natural-language explanations of the injected features. Once trained, the resulting verbalizer explains SAE features directly from decoder directions, addressing both limitations. Our experiments show that the learned verbalization capability generalizes to unseen features, transfers across separately trained SAE dictionaries, and, with a lightweight adapter, extends to SAE features from different LLMs. Intervention experiments show that injecting multiple directions yields an explanation combining their meanings, while reversing individual directions produces corresponding meaning shifts.</description><category>cs.CL</category></item><item><title>Fourier-invariant functions with dense zero sets</title><link>http://arxiv.org/abs/2608.13468</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13468</guid><pubDate>Thu, 13 Aug 2026 16:47:46 +0000</pubDate><dc:creator>Andriy Bondarenko, Kristian Seip</dc:creator><description>Authors: Andriy Bondarenko, Kristian Seip

For every $0\leqβ\leq1/2$, we construct a nonzero real-valued continuous function $f_β$ in $L^1(\mathbb R)\cap L^2(\mathbb R)$ such that $\widehat {f}_β=f_β$ and $f_β(\sqrt{n}/[\log(e+n)]^β)=0$ for all $n\geq 0$. The case $β=0$ settles in the negative a question raised by Radchenko and Viazovska regarding their Fourier interpolation formula. The construction uses a scale of reproducing kernel Hilbert spaces generated by the Fourier-invariant Hermite functions. Applying the Mehler formula, we identify the reproducing kernels of these spaces. By suitable estimates of these kernels, we show that $(\sqrt{n}/[\log(e+n)]^β)$, with one auxiliary point added to it, is a universal interpolating sequence for at least one of the Hilbert spaces under consideration. However, this result fails when $β&gt;1/2$.</description><category>math.CA</category></item><item><title>Deterministic Maximum Likelihood Direction Finding in the Mixture Noise of Gaussian and Spherically Invariant Components</title><link>http://arxiv.org/abs/2608.13294</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13294</guid><pubDate>Thu, 13 Aug 2026 14:26:12 +0000</pubDate><dc:creator>Mingyan Gong</dc:creator><description>Authors: Mingyan Gong

Spherically invariant (SI) random processes can model impulsive noise and unreliable measurements. Recently, the mixture noise of Gaussian and SI components has been used in deterministic maximum likelihood direction finding. In this context, the Expectation-Conditional Maximization (ECM) algorithm, an extension of the expectation-maximization algorithm, has been applied and designed. However, simulation results show that the ECM algorithm always improperly converges. In this article, the ECM Either (ECME) algorithm, an extension of the ECM algorithm, is applied and designed, which additionally utilizes the actual log-likelihood function to first update partial parameter estimates at every iteration and does not need to initialize all parameter estimates. Moreover, the deterministic Cramer-Rao low bounds (CRLBs) of DOA estimators are derived and compared. Simulation results indicate that the ECME algorithm exhibits proper convergence and its root mean square errors of DOA estimates asymptotically approach the CRLBs as the signal powers increase, i.e., the derived CRLBs are correct.</description><category>eess.SP</category><category>stat.AP</category></item><item><title>Block Repetition of Numerical Invariants for the Submodules $[z^k-w^k]$ in $H^2(\mathbb D^2)$</title><link>http://arxiv.org/abs/2608.13275</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13275</guid><pubDate>Thu, 13 Aug 2026 14:10:57 +0000</pubDate><dc:creator>Yin Liu, Yufeng Lu, Chao Zu</dc:creator><description>Authors: Yin Liu, Yufeng Lu, Chao Zu

For $k\ge 2$, let $M_k=[z^k-w^k]$ be the principal homogeneous submodule of the Hardy space over the bidisk. We determine Yang's complete sequence of numerical invariants and prove $$ Σ_0(M_k)=\frac{π^2}{6},\qquad Σ_j(M_k)=Σ_{\lceil j/k\rceil}([z-w]),\quad j\ge1. $$ The proof exploits a residue-class decomposition of the Toeplitz matrices associated with the graded wandering spaces. Consequently, Yang's monotonicity conjecture holds for the family $\{M_k:k\ge2\}$. We also show that the nonzero spectral data of the core operator are independent of $k$, whereas the numerical invariant sequence recovers $k$ from the length of its constant blocks. Thus the higher numerical invariants detect module-theoretic information invisible to the core spectrum.</description><category>math.FA</category><category>math.CV</category></item><item><title>Reconstructing Turaev--Viro Invariants from Four-Fold Simple Branched Covering Presentations</title><link>http://arxiv.org/abs/2608.13241</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13241</guid><pubDate>Thu, 13 Aug 2026 13:45:35 +0000</pubDate><dc:creator>Eri Hatakenaka</dc:creator><description>Authors: Eri Hatakenaka

We construct a finite state sum from a link diagram labelled by transpositions. The diagram presents a closed 3-manifold as a four-fold simple branched cover, and the state sum equals the SU(2)-type Turaev--Viro invariant of that manifold. For each admissible colouring of the diagram's edges and complementary regions, a global factor is combined with crossing weights obtained by summing over compatible local colourings. We then sum these products over all diagram colourings. The formula is proved by assembling crossing-wise local triangulations and matching the two state sums.</description><category>math.GT</category></item><item><title>Infinitely many elliptic curves over $\mathbb{Q}(i)$ of exact ranks 4 and 6 with $j$-invariant 1728</title><link>http://arxiv.org/abs/2608.13202</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13202</guid><pubDate>Thu, 13 Aug 2026 13:05:34 +0000</pubDate><dc:creator>Ben Savoie</dc:creator><description>Authors: Ben Savoie

For each $r\in\{4,6\}$, we construct an explicit one-parameter family of elliptic curves over $\mathbb{Q}(i)$ containing infinitely many pairwise nonisomorphic curves genuinely defined over $\mathbb{Q}(i)$ with $j$-invariant $1728$ and rank exactly $r$. We construct explicit $\mathbb{Q}(i)$-rational points to bound the ranks from below. Kai's theorem on prime values of linear patterns over number fields provides specializations with controlled local behavior, allowing us to obtain matching upper bounds via $[1+i]$-descent. The construction extends the strategy of the author's earlier rank-$2$ paper by replacing a symmetric Gaussian-prime configuration with systems of binary linear forms satisfying several complementary square identities. In the rank-$6$ case, the support vectors attached to the three constructed points and $(0,0)$ span the self-dual Reed-Muller code $\mathrm{RM}(1,3)$, which also occurs as the kernel of the quadratic-residue Laplacian governing the Selmer group.</description><category>math.NT</category></item><item><title>Sinkhorn Linearization and the Spectral Proxy: Unifying the Statistical and Algorithmic Theory of Feature-Parameterized Inverse Optimal Transport via a Single Spectral Sandwich</title><link>http://arxiv.org/abs/2608.13201</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13201</guid><pubDate>Thu, 13 Aug 2026 13:03:55 +0000</pubDate><dc:creator>Han Dong, Jiaming Li, Yongqiang Gong, Ruixi Li, Yin Liu</dc:creator><description>Authors: Han Dong, Jiaming Li, Yongqiang Gong, Ruixi Li, Yin Liu

We develop the statistical and algorithmic theory of inverse optimal transport (IOT) under the feature-parameterized cost C_theta(i,j) = -theta^T phi(i,j). The core technical contribution is the Sinkhorn linearization -- the implicit-function sensitivity of the entropic OT plan to the cost -- together with its spectral proxy, a formula that is spectrally exact yet geometrically transparent. The restricted Hessian on the tangent space satisfies the spectral sandwich (pi_min/epsilon) I &lt;= H_T^{-1} &lt;= (pi_max/epsilon) I, yielding the single core bound sigma_min &gt;= (pi_min/(a_max epsilon)) sqrt(lambda_min(Sigma)) that drives the entire theory. On this core we establish four theorems and one observation. T1 (identifiability): theta is globally injective on the quotient of the gauge kernel, with dimension bound F &lt;= (K-1)^2. T2 (sparsistency): the l1-penalized estimator recovers the true support under irrepresentability and score concentration, with exponential failure probability. T3 (well-posedness): the feature-moment map M(theta) = Phi^T x_theta is strongly monotone, and the inverse is Lipschitz with constant L &lt;= epsilon ||Phi^T S_a||_op / (pi_min lambda_min(Sigma)). T4 (convergence): local strong convexity with mu &gt;= pi_min^2 lambda_min(Sigma) / epsilon^2 guarantees monotone gradient descent convergence. O5 (misspecification): the estimator converges to the OT-model projection of the truth; the Holder continuity of the projection map is assessed numerically, yielding setting-dependent empirical exponents alpha_eff in (0,1).</description><category>stat.ML</category><category>cs.LG</category><category>math.OC</category><category>math.ST</category></item><item><title>Smart Contract Invariants Protect Against Cybercriminals</title><link>http://arxiv.org/abs/2608.13191</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13191</guid><pubDate>Thu, 13 Aug 2026 12:58:36 +0000</pubDate><dc:creator>Sofia Bobadilla, Humaira Afrin, Angela Novelli, Martin Monperrus</dc:creator><description>Authors: Sofia Bobadilla, Humaira Afrin, Angela Novelli, Martin Monperrus

Blockchains are among the most adversarial environments in computing. Billions are stolen by cybercriminals who exploit vulnerabilities. This is an open problem and no concept or technique has proven to really make a difference. In this paper, we claim that the classical notion of program invariant is perhaps the most powerful solution to the problem. We devise anoriginal experimental protocol to 1) study how invariants would have protected against past real-world attacks and 2) whether state-of-the-art automated tools can find them. The experimental toolchain is sophisticated. It is based on INVARIANTEVAL, a benchmark of 28 real Ethereum exploits, each paired with a human-authored invariant that blocks the attack. We validate every invariant with PONDEREPLAY, a replay framework that re-executes transactions in order to prove the correctness and soundness of smart contract invariants. We demonstrate that smart contract invariants block all the cybercriminal attacks in INVARIANTEVAL, fully validated by replaying 108,637 historical transactions. Our large-scale experiments clearly demonstrate that smart contract invariants protect against cybercriminals.</description><category>cs.CR</category><category>cs.SE</category></item><item><title>Tropical and algebraic elliptic plane curves with fixed j-invariant</title><link>http://arxiv.org/abs/2608.13081</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13081</guid><pubDate>Thu, 13 Aug 2026 10:52:20 +0000</pubDate><dc:creator>Alessio Cela, Sae Koyama</dc:creator><description>Authors: Alessio Cela, Sae Koyama

We tropically enumerate planar well-spaced elliptic curves with a fixed $j$-invariant. Combined with the recent genus 1 correspondence theorem of Cela--Koyama, this yields an alternative proof of Pandharipande's algebraic enumeration.</description><category>math.AG</category></item><item><title>On the global feature importance for interpretable and trustworthy heat demand forecasting</title><link>http://arxiv.org/abs/2608.13039</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13039</guid><pubDate>Thu, 13 Aug 2026 10:05:25 +0000</pubDate><dc:creator>Milan Zdravković</dc:creator><description>Authors: Milan Zdravković

The paper introduces the ante-hoc Explainable AI methodology to assess the global feature importance of the Machine Learning models used for heat demand forecasting in intelligent control of District Heating Systems, with motivation to facilitate their interpretability and trustworthiness, hence addressing the challenges related to adherence to communal standards, customer satisfaction and liability risks. Methodology includes use of four different approaches, namely intrinsic interpretability of Gradient Boosting method and selected post-hoc methods, namely Partial Dependence, Accumulated Local Effects and SHAP. None of the selected methods assume feature permutation or perturbations which can introduce bias due to introduction of random unrealistic values of data instances. Discussion of results is provided, including the assessment of complementarities where applicable, with specific interpretations in context of the district heating processes.</description><category>cs.LG</category><category>eess.SY</category></item><item><title>Non-uniqueness of Brakke flows starting from minimal surfaces with singularities</title><link>http://arxiv.org/abs/2608.13531</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13531</guid><pubDate>Thu, 13 Aug 2026 17:48:51 +0000</pubDate><dc:creator>Kotaro Motegi</dc:creator><description>Authors: Kotaro Motegi

We prove the existence of a genuinely time-dependent Brakke flow starting from $Γ_0 \subset \mathbb{R}^{n+1}$ whose associated multiplicity-one varifold is stationary, provided that, at some singular point, the scale-invariant $L^2$ distance of $Γ_0$ from an $n$-dimensional plane has sufficiently small limsup as the scale tends to zero. This yields the dynamical instability of $Γ_0$, a notion recently introduced by Stuvard and Tonegawa, and hence the non-uniqueness of Brakke flows starting from $Γ_0$. A notable feature of our result is that it holds without assuming the uniqueness of tangent cones at the singular point.</description><category>math.AP</category><category>math.DG</category></item><item><title>A pion-driven near-threshold enhancement in the $π\,T_{cc}$ system</title><link>http://arxiv.org/abs/2608.13525</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13525</guid><pubDate>Thu, 13 Aug 2026 17:43:57 +0000</pubDate><dc:creator>Pedro Brandão, Jorgivan Morais Dias, Luciano M. Abreu</dc:creator><description>Authors: Pedro Brandão, Jorgivan Morais Dias, Luciano M. Abreu

Hadronic molecules are commonly viewed as products of near-threshold two-body dynamics. We investigate whether an experimentally established molecule can also serve as a building block for a more complex exotic hadron by studying pion scattering off $T^+_{cc}(3875)$. Treating $T_{cc}$ as a correlated isoscalar $D D^*$ molecular cluster, we describe the resulting $πD D^*$ dynamics in the $I(J^P)=1(1^-)$ channel within the Fixed Center Approximation to the Faddeev equations. The full three-body amplitude develops a pronounced enhancement slightly above the $πT_{cc}$ threshold, centered at $M_{\rm peak}\simeq 4055$~MeV, with an apparent width of approximately $60$~MeV. The associated behavior of the real and imaginary parts of the amplitude is compatible with a resonant-like interpretation. These results show that the lightest hadron can play an active dynamical role in promoting an observed hadronic molecule into a building block of a heavier near-threshold structure generated by the three-body dynamics. We propose searching for this enhancement in the $T_{cc}π$ invariant-mass distribution, particularly through the $D^0D^0π^+π^-$ final state.</description><category>hep-ph</category></item><item><title>Attention from Action, for Action: Emergent Visual Bottlenecks for Policy Learning</title><link>http://arxiv.org/abs/2608.13422</link><guid isPermaLink="true">http://arxiv.org/abs/2608.13422</guid><pubDate>Thu, 13 Aug 2026 16:14:07 +0000</pubDate><dc:creator>Zheyu Zhuang, Ruiyu Wang, Nick Heppert, Johannes Fabian Hahn, Abhinav Valada, Florian T. Pokorny, Danica Kragic</dc:creator><description>Authors: Zheyu Zhuang, Ruiyu Wang, Nick Heppert, Johannes Fabian Hahn, Abhinav Valada, Florian T. Pokorny, Danica Kragic

Visual bottlenecks that focus policy inputs on regions of interest (ROIs) can improve data-efficient visuomotor learning by separating where to look from how to act. Many ROI interfaces rely on external spatial labels, such as gaze, object classes, or affordance annotations. Label-free alternatives often derive crops from trajectories by detecting gripper or motion events and centering a fixed crop at the projected end-effector. Such action-derived crops are useful spatial priors that require no additional labels, but they encode fixed choices about event timing, proxy points, and crop scale. When the visual evidence needed for control lies away from the end-effector or changes continuously with task progress, these crops can become misaligned. We propose Seeker, a task- and state-conditioned readout that learns attention from action. Starting from frozen DINOv3 features, Seeker iteratively updates a query with gathered visual evidence, producing progression-aware ROIs solely from action supervision. The learned ROI serves as a spatial interface for RGB cropping, mask-guided background augmentation, and point-cloud filtering. In simulation and the real world, Seeker improves data efficiency and robustness over no-crop, augmentation, and action-derived crop baselines. On real robots, Seeker raises average in-domain success from the best baseline's 48.3% to 76.7% and success under lighting/background shifts from 20.0% to 60.0%.</description><category>cs.RO</category></item></channel></rss>