Zobrazeno 1 - 10
of 4 786
pro vyhledávání: '"Singhi, A."'
Large Language Models (LLMs) face significant challenges at inference time due to their high computational demands. To address this, we present Performance-Guided Knowledge Distillation (PGKD), a cost-effective and high-throughput solution for produc
Externí odkaz:
http://arxiv.org/abs/2411.05045
Autor:
Binz, Marcel, Akata, Elif, Bethge, Matthias, Brändle, Franziska, Callaway, Fred, Coda-Forno, Julian, Dayan, Peter, Demircan, Can, Eckstein, Maria K., Éltető, Noémi, Griffiths, Thomas L., Haridi, Susanne, Jagadish, Akshay K., Ji-An, Li, Kipnis, Alexander, Kumar, Sreejan, Ludwig, Tobias, Mathony, Marvin, Mattar, Marcelo, Modirshanechi, Alireza, Nath, Surabhi S., Peterson, Joshua C., Rmus, Milena, Russek, Evan M., Saanum, Tankred, Scharfenberg, Natalia, Schubert, Johannes A., Buschoff, Luca M. Schulze, Singhi, Nishad, Sui, Xin, Thalmann, Mirko, Theis, Fabian, Truong, Vuong, Udandarao, Vishaal, Voudouris, Konstantinos, Wilson, Robert, Witte, Kristin, Wu, Shuchen, Wulff, Dirk, Xiong, Huadong, Schulz, Eric
Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in
Externí odkaz:
http://arxiv.org/abs/2410.20268
Autor:
Swain, Swadesh, Singhi, Shree
Integrated Gradients (IG) is a widely used algorithm for attributing the outputs of a deep neural network to its input features. Due to the absence of closed-form integrals for deep learning models, inaccurate Riemann Sum approximations are used to c
Externí odkaz:
http://arxiv.org/abs/2410.04118
Autor:
Singhi, Shree, Kumari, Anupriya
We conducted a reproducibility study on Integrated Gradients (IG) based methods and the Important Direction Gradient Integration (IDGI) framework. IDGI eliminates the explanation noise in each step of the computation of IG-based methods that use the
Externí odkaz:
http://arxiv.org/abs/2409.09043
Autor:
Singhi, Kaustubh
We study certain higher point thermal correlators of heavy and light scalar primaries in a holographic CFT. Assuming simple self-interactions and couplings of the scalars in the bulk theory, we show that the thermal correlators contain a signature of
Externí odkaz:
http://arxiv.org/abs/2406.08553
Concept Bottleneck Models (CBMs) ground image classification on human-understandable concepts to allow for interpretable model decisions. Crucially, the CBM design inherently allows for human interventions, in which expert users are given the ability
Externí odkaz:
http://arxiv.org/abs/2405.01531
Multimodal contrastive pretraining has been used to train multimodal representation models, such as CLIP, on large amounts of paired image-text data. However, previous studies have revealed that such models are vulnerable to backdoor attacks. Specifi
Externí odkaz:
http://arxiv.org/abs/2303.03323
Autor:
Bogdan Kochetov, Phoenix D. Bell, Paulo S. Garcia, Akram S. Shalaby, Rebecca Raphael, Benjamin Raymond, Brian J. Leibowitz, Karen Schoedel, Rhonda M. Brand, Randall E. Brand, Jian Yu, Lin Zhang, Brenda Diergaarde, Robert E. Schoen, Aatur Singhi, Shikhar Uttam
Publikováno v:
Communications Biology, Vol 7, Iss 1, Pp 1-14 (2024)
Abstract Multiplexed imaging technologies have made it possible to interrogate complex tissue microenvironments at sub-cellular resolution within their native spatial context. However, proper quantification of this complexity requires the ability to
Externí odkaz:
https://doaj.org/article/874dc85ae11740acb138f7e5d23648d4
Autor:
Morgan Ferrell, Deniz Can Guven, Cyndi Gonzalez Gomez, Elham Nasrollahi, Richard Giza, Svea Cheng, Masood Pasha Syed, Tara Magge, Aatur Singhi, Anwaar Saeed, Turcin Saridogan, Ibrahim Halil Sahin
Publikováno v:
Scientific Reports, Vol 14, Iss 1, Pp 1-7 (2024)
Abstract Colorectal cancer (CRC) is the third most common cancer in the United States. Recent epidemiological evidence demonstrates an increasing incidence of young-onset CRC cases, defined as CRC cases in individuals 50 years old or younger. Studies
Externí odkaz:
https://doaj.org/article/08470d9b0720485da7e155fe30a333a6
People are often confronted with problems whose complexity exceeds their cognitive capacities. To deal with this complexity, individuals and managers can break complex problems down into a series of subgoals. Which subgoals are most effective depends
Externí odkaz:
http://arxiv.org/abs/2302.02633