Search arXiv⌕ Search

arXiv subjects

Ashish Kumar Singh

Publications and source records attributed to Ashish Kumar Singh.

12 recordsLinked to original sources

MATA: Motion-aligned Temporal Adapter

Few-shot fine-grained video action recognition requires recognizing novel action classes from only a few labeled examples. The difficulty is that visually similar actions often share actors, scenes, poses, and objects, differing only in the order, timing, and short duration of motion changes. We study this problem under three practical constraints: few tunable parameters, low compute cost, and use of the video modality alone. We propose Motion-aligned Temporal Adapter (MATA), a parameter-efficient adapter on top of a frozen CLIP backbone. MATA specializes CLIP features along the temporal dimension using two components: temporal phase alignment through learnable temporal positional embeddings, and a local motion transition operator through depth-wise convolution. With only ~3.6M tunable parameters, MATA achieves state-of-the-art performance across multiple fine-grained benchmarks while remaining robust to both visual corruptions and temporal disruptions induced by frame dropping. Code is available at https://github.com/prakashchhipa/MATA-Motion-aligned-Temporal-Adapter

cs.CV↗

Trust-Limited Machine-Learning-Accelerated Nudged Elastic Band Calculations for Small-Polaron Hopping in Rutile TiO2

Nudged elastic band (NEB) calculations of small-polaron hopping are demanding because nuclear motion is coupled to a self-consistent localized electronic state. Near the charge-transfer region, slow self-consistent-field convergence, localization switching, and competing electronic solutions complicate surrogate-assisted optimization. We develop a Gaussian-process-assisted NEB workflow for this regime and demonstrate electron small-polaron hopping in rutile TiO2 using spin-polarized density functional theory with a Hubbard U correction (DFT+U ) with SIESTA. A joint energy-force Gaussian process (GP) with a Matérn covariance model advances the elastic band, while active-atom root-mean-square and maximum-atom displacement limits constrain surrogate motion relative to evaluated configurations. Reaching these limits triggers first-principles sampling without requiring full surrogate convergence. Three independently DFT-verified 96-atom [001] calculations using 11, 13, and 15 images and different surrogate settings and GP parameterizations yield barriers of 0.1645, 0.1637, and 0.1640 eV, respectively, spanning 0.793 meV. In the 15-image calculation, repeated central-image sampling activates a separately calibrated local GP during surrogate optimization. Final full-band DFT verification gives a maximum projected NEB force of 0.0346 eV/Å. A 324-atom calculation yields a lower barrier of 0.1302 eV, indicating supercell sensitivity and demonstrating feasibility at larger GP observation dimension. The workflow combines controlled surrogate extrapolation, targeted charge-transfer-region interpolation, and independent DFT verification to obtain polaron migration pathways, emphasizing reproducible barriers and full-band force convergence.

cond-mat.mtrl-sci↗

DiA: Directional Adapter

Fine-grained action recognition often depends on how an action unfolds over time rather than on appearance alone. Visually similar classes may differ in preparation, execution, or completion phases, making uniformly aggregated temporal context suboptimal. We study this problem under a practical objective: i) improving recognition performance with few tunable parameters, ii) being compute efficient, and iii) using only the readily available video modality. Motivated by this missing temporal observation and the accuracy--compute efficiency--unimodality objective, we propose Directional Adapter (DiA), a parameter-efficient adapter on top of CLIP. DiA formulates temporal specialization into causal and anti-causal directions, allowing past-to-present and future-to-present cues to be modeled differently. To keep this directional modeling parameter- and compute-efficient, both directions use depth-wise temporal convolutions in a compact bottleneck space and are combined through a learnable fusion. DiA achieves state-of-the-art performance among methods using a single video input modality across seven public benchmarks, with only $\sim$3.6M tunable parameters, while also delivering higher throughput and lower inference latency than prior methods. Code is available at https://github.com/prakashchhipa/DiA-Directional-Adapter.

cs.CV↗

Telescopic Language Models

One deployed language model must often serve many compute budgets, yet serving each budget still means a separate training or compression run per point. We train a Telescopic Language Model (TLM) to be that continuum: a nested-capacity Transformer supervised by stochastic prefix supervision with a full anchor. At every step, one randomly truncated prefix of the capacity axis is trained against the full next-token target, alongside one full-capacity pass, so the trained artifact is a valid language model at every depth. Two forward-backward passes per step, no architectural change, nothing extra at inference. Fixed-exit suites such as Matryoshka Language Model Suites (MLMS) occupy one point in this design space, and the point has a cost: supervising only a few fixed exits leaves the nested model at chance level everywhere else (perplexity 10^2-10^5 in our baselines). On a 200M proxy suite (20B FineWeb-Edu tokens, identical data stream for all methods), a single TLM run is a valid language model at every one of its twenty layer prefixes, in perplexity and on perplexity-sensitive downstream tasks, reducing the area under the quality-budget curve by 43-44% relative to the fixed-exit suites while matching them at full capacity, at ~12% lower GPU cost per run. The prefix sampling density is a dial: concentrating it on a few depths recovers fixed-exit quality there at the price of the continuum, so the operating points become a training-time choice rather than an architectural one. These results indicate that the training objective, not the nesting itself, is what makes a model elastic.

cs.CL↗

Mesoscopic simulations of linear and ring polymer solutions with explicit hydrodynamics under good and poor solvent conditions

We employ large-scale Dissipative Particle Dynamics simulations to investigate dilute solutions of linear polymers and unknotted, non-concatenated ring polymers in explicit solvent. By systematically varying solvent quality, we examine the interplay between hydrodynamic interactions, chain architecture, and intermolecular association. Under good solvent conditions, both linear and ring polymers remain expanded and well dispersed, displaying center-of-mass dynamics consistent with normal diffusion. In poor solvents, attractive polymer-polymer interactions drive the formation of irregular aggregates characterized by partial chain collapse, substantial interpenetration, and slower dynamics. Despite their different topologies, the two polymer architectures exhibit remarkably similar structural and dynamical responses across the solvent conditions considered. These results indicate that solvent quality largely determines the organization and transport properties of dilute polymer solutions, whereas topological effects remain comparatively weak in the investigated regime.

cond-mat.soft↗

On the Codesign of Scientific Experiments and Industrial Systems

The optimization of large experiments in fundamental science, such as detectors for subnuclear physics at particle colliders, shares with the optimization of complex systems for industrial or societal applications the common issue of addressing the inter-relation between parameters describing the hardware used in data production and parameters used to analyse those data. While in many cases this coupling can be ignored -- when the problem can be successfully factored into simpler sub-tasks and the latter addressed serially -- there are situations in which that approach fails to converge to the absolute maximum of expected performance, as it results in a mis-alignment of the optimized hardware and software solutions. In this work we consider a few use cases of interest in fundamental science collected primarily from particle physics and related areas, and a pot-pourri of industrial and societal applications where the matter is similarly of relevance. We discuss the emergence of strong hardware-software coupling in some of those systems, as well as co-design procedures that may be deployed to identify the global maximum of their relevant utility functions. We observe how numerous opportunities exist to advance methods and tools for hardware-software co-design optimization, bridging fundamental science and industry through application- and challenge-driven projects, and shaping the future of scientific experiments and industrial systems.

physics.ins-det↗

Restereo: Diffusion stereo video generation and restoration

Stereo video generation has been gaining increasing attention with recent advancements in video diffusion models. However, most existing methods focus on generating 3D stereoscopic videos from monocular 2D videos. These approaches typically assume that the input monocular video is of high quality, making the task primarily about inpainting occluded regions in the warped video while preserving disoccluded areas. In this paper, we introduce a new pipeline that not only generates stereo videos but also enhances both left-view and right-view videos consistently with a single model. Our approach achieves this by fine-tuning the model on degraded data for restoration, as well as conditioning the model on warped masks for consistent stereo generation. As a result, our method can be fine-tuned on a relatively small synthetic stereo video datasets and applied to low-quality real-world videos, performing both stereo video generation and restoration. Experiments demonstrate that our method outperforms existing approaches both qualitatively and quantitatively in stereo video generation from low-resolution inputs.

cs.CV↗

Enhancing Cross-Category Learning in Recommendation Systems with Multi-Layer Embedding Training

Modern DNN-based recommendation systems rely on training-derived embeddings of sparse features. Input sparsity makes obtaining high-quality embeddings for rarely-occurring categories harder as their representations are updated infrequently. We demonstrate a training-time technique to produce superior embeddings via effective cross-category learning and theoretically explain its surprising effectiveness. The scheme, termed the multi-layer embeddings training (MLET), trains embeddings using factorization of the embedding layer, with an inner dimension higher than the target embedding dimension. For inference efficiency, MLET converts the trained two-layer embedding into a single-layer one thus keeping inference-time model size unchanged. Empirical superiority of MLET is puzzling as its search space is not larger than that of the single-layer embedding. The strong dependence of MLET on the inner dimension is even more surprising. We develop a theory that explains both of these behaviors by showing that MLET creates an adaptive update mechanism modulated by the singular vectors of embeddings. When tested on multiple state-of-the-art recommendation models for click-through rate (CTR) prediction tasks, MLET consistently produces better models, especially for rare items. At constant model quality, MLET allows embedding dimension, and model size, reduction by up to 16x, and 5.8x on average, across the models.

cs.LG↗

Phase separation kinetics of block copolymer melts confined under moving parallel walls: a DPD study

We use dissipative particle dynamics (DPD) simulations to study the effect of shear on domain morphology and kinetics of microphase separating critical diblock copolymer (BCP) bulk melts. The melt is confined within two parallel solid walls at the top and bottom of the simulation box. The shear is induced by allowing the walls to move in a direction with a specific velocity. We explore the following cases: (i) walls are fixed, (ii) only the top wall moves, (iii) both walls move in the same direction, and (iv) both walls move in opposite directions. After the temperature quench, we monitor the effect of shear on evolution morphology, the scaling behavior of the system, and the characteristic length scale and growth. The characteristic length scale follows typical power-law behavior at early times and saturates at late times when both walls are fixed. The length scale changes significantly with shear caused by wall velocities. The usual lamellar morphology, which is not achieved for case 1 within the considered simulation time steps, is noticed much earlier for the nonzero wall velocity cases. Specifically, it is seen much before in case 4 than in the other cases. We find that the shear viscosity decreases (shear-thinning) with wall velocity (shear rate) for all the cases at a given coarsening time. Overall, we report the influence rule of shear rates on microphase separation kinetics of BCP melts. This study can provide a scheme to anticipate and design anisotropic microstructures under the application of externally controlled wall shear that may further guide in producing the various composite materials with superior mechanical and physical properties.

cond-mat.soft↗

Photo-induced bond breaking during phase separation kinetics of block copolymer melts: A dissipative particle dynamics study

Using dissipative particle dynamics (DPD) simulation method, we study the phase separation dynamics in block copolymer (BCP) melt in $d=3$, subjected to external stimuli such as light. An initial homogeneous BCP melt is rapidly quenched to a temperature $T < T_c$, where $T_c$ is the critical temperature. We then let the system go through alternate light "on" and "off" cycles. An on-cycle breaks the stimuli-sensitive bonds connecting both the blocks A and B in BCP melt, and during the off-cycle, broken bonds reconnect. By simulating the effect of light, we isolate scenarios where phase separation begins with the light off (set 1); the cooperative interactions within the system allow it to undergo microphase separation. When the phase separation starts with the light on (set 2), the system undergoes macrophase separation due to the bond breaking. Here, we report the role of alternate cycles on domain morphology by varying bond-breaking probability for both the sets 1 and 2, respectively. We observe that the scaling functions depend upon the conditions mentioned above that change the time scale of the evolving morphologies in various cycles. However, in all the cases, the average domain size respects the power-law growth: $R(t)\sim t^ϕ$ at late times, here $ϕ$ is the dynamic growth exponent. After a short-lived diffusive growth ($ϕ\sim 1/3$) at early times, $ϕ$ illustrates a crossover from the viscous hydrodynamic ($ϕ\sim 1$) to the inertial hydrodynamic ($ϕ\sim 2/3$) regimes at late times.

cond-mat.soft↗

Training with Multi-Layer Embeddings for Model Reduction

Modern recommendation systems rely on real-valued embeddings of categorical features. Increasing the dimension of embedding vectors improves model accuracy but comes at a high cost to model size. We introduce a multi-layer embedding training (MLET) architecture that trains embeddings via a sequence of linear layers to derive superior embedding accuracy vs. model size trade-off. Our approach is fundamentally based on the ability of factorized linear layers to produce superior embeddings to that of a single linear layer. We focus on the analysis and implementation of a two-layer scheme. Harnessing the recent results in dynamics of backpropagation in linear neural networks, we explain the ability to get superior multi-layer embeddings via their tendency to have lower effective rank. We show that substantial advantages are obtained in the regime where the width of the hidden layer is much larger than that of the final embedding (d). Crucially, at conclusion of training, we convert the two-layer solution into a single-layer one: as a result, the inference-time model size scales as d. We prototype the MLET scheme within Facebook's PyTorch-based open-source Deep Learning Recommendation Model. We show that it allows reducing d by 4-8X, with a corresponding improvement in memory footprint, at given model accuracy. The experiments are run on two publicly available click-through-rate prediction benchmarks (Criteo-Kaggle and Avazu). The runtime cost of MLET is 25%, on average.

cs.LG↗

Generation of Efficient Codes for Realizing Boolean Functions in Nanotechnologies

We address the challenge of implementing reliable computation of Boolean functions in future nanocircuit fabrics. Such fabrics are projected to have very high defect rates. We overcome this limitation by using a combination of cheap but unreliable nanodevices and reliable but expensive CMOS devices. In our approach, defect tolerance is achieved through a novel coding of Boolean functions; specifically, we exploit the dont cares of Boolean functions encountered in multi-level Boolean logic networks for constructing better codes. We show that compared to direct application of existing coding techniques, the coding overhead in terms of extra bits can be reduced, on average by 23%, and savings can go up to 34%. We demonstrate that by incorporating efficient coding techniques more than a 40% average yield improvement is possible in case of 1% and 0.1% defect rates. With 0.1% defect density, the savings can be up to 90%.

cs.IT↗