Searcharxiv⌕ Search

arXiv subjects

Lakpa Tamang

Publications and source records attributed to Lakpa Tamang.

8 recordsLinked to original sources

Beyond Raw Similarity: Prior-Aware Fusion for Out-of-Distribution Detection

Out-of-distribution (OOD) detection in vision-language models presents unique challenges due to the reliance on cross-modal similarity. Even when embeddings are well separated for in-distribution (ID) data, similarity scores can remain overconfident for OOD inputs. A common remedy is to adapt the model on ID data, but adaptation discards the pretrained geometry that made the similarity meaningful. We propose multi-modal logit fusion (MM-Log), which keeps both by pairing a gradient-free zero-shot prior with a task-adapted branch in each modality and fuses their four cross-modal similarity channels into a single OOD score. Because the zero-shot priors are never adapted, the cross-channels act as a consistency check between the pretrained and adapted geometries: an OOD input that fools the adapted branch through a shortcut must additionally agree with frozen CLIP to retain a high score. We formalize this as a margin bound showing that the fused score is provably penalized by disagreement between the zero-shot and adapted branches, a signal to which adapted-only scores are structurally blind. MM-Log requires no auxiliary OOD data, negative prompts, or test-time adaptation, and trains on few-shot ID data alone. Across Far-OOD, Near-OOD, and Hard-OOD benchmarks, MM-Log consistently outperforms strong CLIP-based baselines.

cs.CV↗

Covariance-aware Scoring: A Geometric Correction for Vision-Language Test-Time Adaptation

Test-time adaptation (TTA) enables vision--language models such as CLIP to adapt to distribution shifts without labels or access to source data. Existing methods adapt through feature caching, prototype estimation, or the online accumulation of feature statistics, but most ultimately score a test feature against a class representative in the ambient embedding space, typically through cosine similarity. We identify this shared decision rule as a limitation under anisotropic feature distributions, which means when a small number of directions account for a large fraction of feature variance, the score can be dominated by high-variance directions that are not necessarily discriminative. In light of this, we propose CARe, a covariance-aware scoring for TTA that replaces the ambient geometry with a Mahalanobis geometry induced by the target feature second moment. Theoretically, we show that under a shared-covariance Gaussian model the covariance-aware decision direction coincides with that of the Bayes-optimal linear discriminant, with ambient-space scoring as the isotropic special case, and establish that the potential benefit of covariance-aware scoring grows with feature anisotropy. Empirically, we show that CARe improves adaptation accuracy over its base cosine scoring on both standard and fine-grained distribution-shift benchmarks across two backbones. Consistent with our analysis, the gains are largest on datasets whose features are most anisotropic and vanish as the features approach isotropy.

cs.CV↗

Spin-valley physics in anomalous thermoelectric responses of the spin-orbit coupled $α$-$T_3$ system with broken time-reversal symmetry

We extract spin-valley physics in the anomalous Hall and Nernst responses of the spin-orbit coupled $α$-$T_3$ system in the presence of a time-reversal symmetry breaking staggered magnetization. We show that the interplay between the SOI, magnetization, and a model parameter $α$ for the $α$-$T_3$ lattice enables efficient tuning of spin- and valley-dependent Hall and Nernst signals. The spin-valley physics of the Hall and Nernst responses in the absence and presence of the magnetization are well explained. The peak-dip features of the Nernst responses are also understood from the corresponding Hall responses through the Mott relation. We find that the magnetization introduces highly tunable spin and valley polarizations, which are calculated from the spin- and valley-resolved Nernst conductivities. It is shown that both the spin and valley polarizations can attain nearly complete polarization over extended regions of the parameter space.

cond-mat.mes-hall↗

DMS2F-HAD: A Dual-branch Mamba-based Spatial-Spectral Fusion Network for Hyperspectral Anomaly Detection

Hyperspectral anomaly detection (HAD) aims to identify rare and irregular targets in high-dimensional hyperspectral images (HSIs), which are often noisy and unlabelled data. Existing deep learning methods either fail to capture long-range spectral dependencies (e.g., convolutional neural networks) or suffer from high computational cost (e.g., Transformers). To address these challenges, we propose DMS2F-HAD, a novel dual-branch Mamba-based model. Our architecture utilizes Mamba's linear-time modeling to efficiently learn distinct spatial and spectral features in specialized branches, which are then integrated by a dynamic gated fusion mechanism to enhance anomaly localization. Across fourteen benchmark HSI datasets, our proposed DMS2F-HAD not only achieves a state-of-the-art average AUC of 98.78%, but also demonstrates superior efficiency with an inference speed 4.6 times faster than comparable deep learning methods. The results highlight DMS2FHAD's strong generalization and scalability, positioning it as a strong candidate for practical HAD applications.

cs.CV↗

Handling Out-of-Distribution Data: A Survey

In the field of Machine Learning (ML) and data-driven applications, one of the significant challenge is the change in data distribution between the training and deployment stages, commonly known as distribution shift. This paper outlines different mechanisms for handling two main types of distribution shifts: (i) Covariate shift: where the value of features or covariates change between train and test data, and (ii) Concept/Semantic-shift: where model experiences shift in the concept learned during training due to emergence of novel classes in the test phase. We sum up our contributions in three folds. First, we formalize distribution shifts, recite on how the conventional method fails to handle them adequately and urge for a model that can simultaneously perform better in all types of distribution shifts. Second, we discuss why handling distribution shifts is important and provide an extensive review of the methods and techniques that have been developed to detect, measure, and mitigate the effects of these shifts. Third, we discuss the current state of distribution shift handling mechanisms and propose future research directions in this area. Overall, we provide a retrospective synopsis of the literature in the distribution shift, focusing on OOD data that had been overlooked in the existing surveys.

cs.LG↗

Orbital magnetization senses the topological phase transition in a spin-orbit coupled $α$-$T_3$ system

The $α$-$T_3$ system undergoes a topological phase transition(TPT) between two distinct quantum spin-Hall phases across $α=0.5$ when the spin-orbit interaction of Kane-Mele type is taken into consideration. As a hallmark of such a TPT, we find that the Berry curvature and the orbital magnetic moment change their respective signs across the TPT. We also find the trails of the TPT in another physical observable, namely, the orbital magnetization(OM) that can be, in principle, detected experimentally through the circular dichroism associated with optical absorption. The topological features of the OM are understood in terms of valley and spin physics. The valley-resolved OM(VROM) and the spin-resolved OM(SROM) exhibit interesting characteristics related to the valley and the spin Chern number when the chemical potential is tuned in the forbidden gap(s) of the energy spectrum. In particular, we find that the slope of the VROM versus the chemical potential in the forbidden gap changes its sign abruptly across the TPT, which is also consistent with the corresponding change in the valley Chern number. Moreover, the slope of the SROM demonstrates a sudden jump by one unit of $e/h$ (where $e$ is the electronic charge and $h$ is the Planck's constant) across the TPT, which is also in agreement with the corresponding change in the spin Chern number. It is further seen that a definite spin-valley optical selection rule governs the circular dichroism. The $k$-resolved degree of the optical polarization and the low-frequency differential optical absorbance manifest sign change across the TPT. We discuss experimentally viable signatures of different quantum spin-Hall phases in the optical absorbance.

cond-mat.mes-hall↗

Probing Topological signatures in an optically driven $α$-${T_3}$ Lattice

The $α$-$T_3$ lattice, an interpolation model between the honeycomb lattice of graphene($α=0$) and the dice lattice($α=1$), undergoes a topological phase transition across $α=1/\sqrt{2}$ when exposed to a circularly polarized off-resonant light. We study Berry phase mediated bulk magnetic and anomalous thermoelectric responses in order to capture the topological signatures of a driven $α$-$T_3$ lattice. It is revealed that both the Berry curvature and the orbital magnetic moment associated with the flat band change their respective signs across $α=1/\sqrt{2}$. The off-resonant light distorts the flat band near the Dirac points when $0<α<1$ which eventually introduces two distinct well separated forbidden gaps of equal width in the quasienergy spectrum. The orbital magnetization varies linearly with the chemical potential, in the forbidden gaps. The slopes of the linear regions in the orbital magnetization are closely related to the respective Chern numbers on either side of $α=1/\sqrt{2}$. We find that the slope for $α>1/\sqrt{2}$ is approximately two times of that for $α<1/\sqrt{2}$ which essentially indicates a topological phase transition across $α=1/\sqrt{2}$. However, the anomalous Nernst coefficient vanishes when the chemical potential is tuned in the forbidden gaps. The anomalous Hall conductivity in the forbidden gap(s) approaches different quantized values on either side of $α=1/\sqrt{2}$. All these topological signatures can be observed experimentally.

cond-mat.mes-hall↗

Floquet engineering of low-energy dispersions and dynamical localization in a periodically kicked three-band system

Much having learned about Floquet dynamics of pseudospin-$1/2$ system namely, graphene, we here address the stroboscopic properties of a periodically kicked {three-band fermionic system such as $α$-T$_3$ lattice. This particular model provides an interpolation between graphene and dice lattice via the continuous tuning of the parameter $α$ from 0 to 1.} In the case of dice lattice ($α=1$), we reveal that one can, in principle, engineer various types of low energy dispersions around some specific points in the Brillouin zone by tuning the kicking parameter in the Hamiltonian along a particular direction. Our analytical analysis shows that one can experience different quasienergy dispersions for example, Dirac type, semi-Dirac type, gapless line, absolute flat quasienergy bands, depending on the specific values of the kicking parameter. Moreover, we numerically study the dynamics of a wave packet in dice lattice. The quasienergy dispersion allows us to understand the instantaneous structure of wave packet at stroboscopic times. We find a situation where absolute flat quasienergy bands lead to a complete dynamical localization of the wave packet. {Aditionally, we calculate the quasienergy spectrum numerically for $α$-T$_3$ lattice. A periodic kick in a perpendicular (planar) direction breaks (preserves) the particle-hole symmetry for $0<α<1$. Furthermore, it is also revealed that the dynamical localization of wave packet does not occur at any intermediate $α\ne 0,\,1$.}

cond-mat.mes-hall↗