arXiv subjects

Tanvi Sharma

Publications and source records attributed to Tanvi Sharma.

11 recordsLinked to original sources

arXiv2026

NeuroViz: Real-time Interactive Visualization of Forward and Backward Passes in Neural Network Training

Training neural networks is difficult to interpret, particularly for newcomers. We introduce NeuroViz, an interactive visualization tool that supports real-time exploration of fully connected neural network training. Users can configure network architecture, activation functions, learning rates, and datasets, then observe activations, weight updates, and loss progression. NeuroViz visualizes weight changes in direct correspondence with activation signals in both forward and backward passes, enabling users to distinguish pre- and post-update states within individual epochs and view dynamically updating per-neuron equations. We conduct a comparative user study with 31 participants against six established visualization tools and we achieved the highest usability score (SUS 80.97, in the 'excellent' range), with mean rankings of 2.47 for clarity and 2.23 for usefulness (lower is better). Over 70% of participants reported that the visualizations substantially increased their perception of neural network training transparency. The implemented instance is accessible at https://neuroviz.org.

cs.LG↗

arXiv2026

Gaia24ccy: An outburst followed the footsteps of its predecessor

Accretion-driven outbursts in young stellar objects remain poorly understood, largely limited by a statistically small sample of closely followed-up events. This underscores the importance of a thorough exploration of each outbursting object. We studied a peculiar outbursting system, Gaia24ccy, which exhibited two $\Delta g \sim$ 3.8 mag outbursts in 2019 and 2024. The system consists of two unresolved, nearly identical, and rapidly rotating young stars: Gaia24ccy A (1.1419 days) and Gaia24ccy B (1.7898 days). Periodogram analyses just before the onset of the outbursts suggest Gaia24ccy B to be the outbursting component. Unlike any previously known EXor sources, the two outburst profiles show a very similar evolution: both rose at the same rate for the first 15 days, followed by many 'sub-bursts' on the timescale of 10-20 days. The 2019 outburst lasted 145-255 days, while the 2024 outburst persisted for 367 days. We infer the unstable region to lie at $r_{\rm trigger} \simeq 0.019-0.047$ au ($\sim5-12.3 R_\star$). The accreted mass per event $M_{\rm acc}\sim10^{-5} M_\odot$ can be provided by a compact inner-disk reservoir. The photometric rise/decay timescales and the mid-infrared color evolution favor a thermal-viscous trigger in a hot inner disk, while the appearance of rich emission-line spectra indicates concurrent magnetospheric compression - together best described by a hybrid picture. Finally, we explain the reddening of the mid-infrared color observed during the outburst as a consequence of the competing emission from the viscous disk and the photosphere.

astro-ph.SR↗

arXiv2026

More Than a Quick Glance: Overcoming the Greedy Bias in KV-Cache Compression

While Large Language Models (LLMs) can theoretically support extensive context windows, their actual deployment is constrained by the linear growth of Key-Value (KV) cache memory. Prevailing compression strategies mitigate this through various pruning mechanisms, yet trade-off semantic recall for memory efficiency. In this work, we present LASER-KV (Layer Accumulated Selection with Exact-LSH Recall), a framework designed to test the limits of KV compression under a strict accumulative budgeting policy. We deviate from the standard fixed summary size approach by implementing a block-wise accumulation strategy governed by a protection divisor (n). This allows us to isolate the effects of compression from sliding window artifacts. Our experiments on the Babilong benchmark reveal performance degradation in previous compression methods by 15-30% on various long context tasks. LASER-KV maintains stable performance, achieving superior accuracies by a margin of upto 10% at 128k. These findings challenge the prevailing assumption that attention scores alone are a sufficient proxy for token utility.

cs.AI↗

arXiv2025

Hardware-Software Co-Design for Accelerating Transformer Inference Leveraging Compute-in-Memory

Transformers have become the backbone of neural network architecture for most machine learning applications. Their widespread use has resulted in multiple efforts on accelerating attention, the basic building block of transformers. This paper tackles the challenges associated with accelerating attention through a hardware-software co-design approach while leveraging compute-in-memory(CIM) architecture. In particular, our energy- and area-efficient CIM based accelerator, named HASTILY, aims to accelerate softmax computation, an integral operation in attention, and minimize their high on-chip memory requirements that grows quadratically with input sequence length. Our architecture consists of novel CIM units called unified compute and lookup modules(UCLMs) that integrate both lookup and multiply-accumulate functionality within the same SRAM array, incurring minimal area overhead over standard CIM arrays. Designed in TSMC 65nm, UCLMs can be used to concurrently perform exponential and matrix-vector multiplication operations. Complementing the proposed architecture, HASTILY features a fine-grained pipelining strategy for scheduling both attention and feed-forward layers, to reduce the quadratic dependence on sequence length to linear dependence. Further, for fast softmax computation which involves computing the maxima and sum of exponential values, such operations are parallelized across multiple cores using reduce and gather strategy. We evaluate our proposed architecture using a compiler tailored towards attention computation and a standard cycle-level CIM simulator. Our evaluation shows end-to-end throughput(TOPS) improvement of 4.4x-9.8x and 1.7x-5.9x over Nvidia A40 GPU and baseline CIM hardware, respectively, for BERT models with INT-8 precision. Additionally, it shows gains of 16x-36x in energy-efficiency(TOPS/W) over A40 GPU and similar energy-efficiency as baseline CIM hardware.

cs.AR↗

arXiv2024

A Novel Survey for Young Substellar Objects with the W-band Filter. VII. Water-Bearing Objects in the Core of the Rho Ophiuchi Cloud Complex

We present a study of very low-mass stars and brown dwarfs in the rich star-forming core of the Rho Ophiuchi cloud complex. The selection of the sample relies on detecting the inherent water absorption characteristic in young substellar objects. Of the 22 water-bearing candidates selected, 15 have a spectral type of M6 or later. Brown dwarf candidates too faint for membership determination by Gaia have their proper motions derived by deep-infrared images spanning six years. Astrometric analysis confirms 21/22 sources as members, one identified as a contaminant. Infrared colors and the spectral energy distribution of each water-bearing candidate are used to diagnose the mass, age, and possible existence of circumstellar dust. 15 sources exhibit evidence of disks in their spectral energy distributions, as late as in M8-type objects. Spectroscopy for bright candidates has confirmed one as an M8 member and verified two sources (with disks) exhibiting signatures of magnetospheric accretion.

astro-ph.SR↗

arXiv2023

WWW: What, When, Where to Compute-in-Memory

Matrix multiplication is the dominant computation during Machine Learning (ML) inference. To efficiently perform such multiplication operations, Compute-in-memory (CiM) paradigms have emerged as a highly energy efficient solution. However, integrating compute in memory poses key questions, such as 1) What type of CiM to use: Given a multitude of CiM design characteristics, determining their suitability from architecture perspective is needed. 2) When to use CiM: ML inference includes workloads with a variety of memory and compute requirements, making it difficult to identify when CiM is more beneficial than standard processing cores. 3) Where to integrate CiM: Each memory level has different bandwidth and capacity, creating different data reuse opportunities for CiM integration. To answer such questions regarding on-chip CiM integration for accelerating ML workloads, we use an analytical architecture-evaluation methodology with tailored mapping algorithm. The mapping algorithm aims to achieve highest weight reuse and reduced data movements for a given CiM prototype and workload. Our analysis considers the integration of CiM prototypes into the cache levels of a tensor-core-like architecture, and shows that CiM integrated memory improves energy efficiency by up to 3.4x and throughput by up to 15.6x compared to established baseline with INT-8 precision. We believe the proposed work provides insights into what type of CiM to use, and when and where to optimally integrate it in the cache hierarchy for efficient matrix multiplication.

cs.AR↗

arXiv2023

A novel survey for young substellar objects with the W-band filter VI: Spectroscopic census of sub-stellar members and the IMF of $σ$ Orionis cluster

Low-mass stars and sub-stellar objects are essential in tracing the initial mass function (IMF). We study the nearby young $σ$ Orionis cluster (d$\sim$408 pc; age$\sim$1.8 Myr) using deep NIR photometric data in J, W and H-bands from WIRCam on the Canada-France-Hawaii Telescope. We use the water absorption feature to photometrically select the brown dwarfs and confirm their nature spectroscopically with the IRTF-SpeX. Additionally we select candidate low-mass stars for spectroscopy and analyze their membership and that of literature sources using astrometry from Gaia DR3. We obtain the near-IR spectra for 28 very low-mass stars and brown dwarfs and estimate their spectral type between M3-M8.5 (mass ranging between 0.3-0.01 M$_{\odot}$). Apart from these, we also identify 5 new planetary mass candidates which require further spectroscopic confirmation of youth. We compile the comprehensive catalog of 170 spectroscopically confirmed members in the central region of the cluster, for a wide mass range of $\sim$19-0.004 M$_{\odot}$. We estimate the star/BD ratio to be $\sim$4, within the range reported for other nearby star forming regions. With the updated catalog of members we trace the IMF down to 4 M$_\mathrm{Jup}$ and we find that a two-segment power-law fits the sub-stellar IMF better than the log-normal distribution.

astro-ph.SR↗

arXiv2022

A Novel Survey for Young Substellar Objects with the W band Filter.V. IC 348 and Barnard 5 in the Perseus Cloud

We report the discovery of substellar objects in the young star cluster IC 348 and the neighboring Barnard 5 dark cloud, both at the eastern end of the Perseus star-forming complex. The substellar candidates are selected using narrowband imaging, i.e., on and off photometric technique with a filter centered around the water absorption feature at 1.45 microns, a technique proven to be efficient in detecting water-bearing substellar objects. Our spectroscopic observations confirm three brown dwarfs in IC 348. In addition, the source WBIS 03492858+3258064, reported in this work, is the first confirmed brown dwarf discovered toward Barnard 5. Together with the young stellar population selected via near- and mid-infrared colors using the Two Micron All Sky Survey and the Wide-field Infrared Survey Explorer, we diagnose the relation between stellar versus substellar objects with the associated molecular clouds. Analyzed by Gaia EDR3 parallaxes and kinematics of the cloud members across the Perseus region, we propose the star formation scenario of the complex under influence of the nearby OB association.

astro-ph.SR↗

arXiv2022

Diagnosing Triggered Star Formation in the Galactic H II region Sh 2-142

Stars are formed by gravitational collapse, spontaneously or, in some cases under the constructive influence of nearby massive stars, out of molecular cloud cores. Here we present an observational diagnosis of such triggered formation processes in the prominent \ion{H}{2} region Sh\,2-142, which is associated with the young star cluster NGC\,7380, and with some bright-rimmed clouds as the signpost of photoionization of molecular cloud surfaces. Using near- (2MASS) and mid-infrared (WISE) colors, we identified candidate young stars at different evolutionary stages, including embedded infrared sources having spectral energy distributions indicative of active accretion. We have also used data from our optical observations to be used in SEDs, and from Gaia EDR3 to study the kinematics of young objects. With this young stellar sample, together with the latest CO line emission data (spectral resolution $\sim 0.16$~km~s$^{-1}$, sensitivity $\sim 0.5$~K), a positional and ageing sequence relative to the neighboring cloud complex, and to the bright-rimmed clouds, is inferred. The propagating stellar birth may be responsible, at least partially, for the formation of the cluster a few million years ago, and for the ongoing activity now witnessed in the cloud complex.

astro-ph.GA↗

arXiv2021

The JCMT Transient Survey: Four Year Summary of Monitoring the Submillimeter Variability of Protostars

We present the four-year survey results of monthly submillimeter monitoring of eight nearby ($< 500 $pc) star-forming regions by the JCMT Transient Survey. We apply the Lomb-Scargle Periodogram technique to search for and characterize variability on 295 submillimeter peaks brighter than 0.14 Jy beam$^{-1}$, including 22 disk sources (Class II), 83 protostars (Class 0/I), and 190 starless sources. We uncover 18 secular variables, all of them protostars. No single-epoch burst or drop events and no inherently stochastic sources are observed. We classify the secular variables by their timescales into three groups: Periodic, Curved, and Linear. For the Curved and Periodic cases, the detectable fractional amplitude, with respect to mean peak brightness, is $\sim4$ % for sources brighter than $\sim$ 0.5 Jy beam$^{-1}$. Limiting our sample to only these bright sources, the observed variable fraction is 37 % (16 out of 43). Considering source evolution, we find a similar fraction of bright variables for both Class 0 and Class I. Using an empirically motivated conversion from submillimeter variability to variation in mass accretion rate, six sources (7 % of our full sample) are predicted to have years-long accretion events during which the excess mass accreted reaches more than 40 % above the total quiescently accreted mass: two previously known eruptive Class I sources, V1647 Ori and EC 53 (V371 Ser), and four Class 0 sources, HOPS 356, HOPS 373, HOPS 383, and West 40. Considering the full protostellar ensemble, the importance of episodic accretion on few years timescale is negligible, only a few percent of the assembled mass. However, given that this accretion is dominated by events of order the observing time-window, it remains uncertain as to whether the importance of episodic events will continue to rise with decades-long monitoring.

astro-ph.SR↗

arXiv2021

A Novel Survey for Young Substellar Objects with the W-band filter III: Searching for very low mass brown dwarfs in Serpens South and Serpens Core

We present CFHT photometry and IRTF spectroscopy of low-mass candidate members of Serpens South and Serpens Core ($\sim$430 pc, $\sim$0.5 Myr), identified using a novel combination of photometric filters, known as the W-band method. We report SC182952+011618, SS182959-020335 and SS183032-021028 as young, low-mass Serpens candidate members, with spectral types in the range M7-M8, M5-L0 and M5-M6.5 respectively. Best-fit effective temperatures and luminosities imply masses of $<$ 0.12M$_{\odot}$ for all three candidate cluster members. We also present Hubble Space Telescope imaging data (F127M, F139M and F850LP) for six targets in Serpens South. We report the discovery of the binary system SS183044-020918AB. The binary components are separated by $\approx$45 AU, with spectral types of M7-M8 and M8-M9, and masses of 0.08-0.1 and 0.05-0.07M$_{\odot}$. We discuss the effects of high dust attenuation on the reliability of our analysis, as well as the presence of reddened background stars in our photometric sample.

astro-ph.SR↗