Zobrazeno 1 - 10
of 230
pro vyhledávání: '"Kumaran, Dharshan"'
Autor:
Barbero, Federico, Banino, Andrea, Kapturowski, Steven, Kumaran, Dharshan, Araújo, João G. M., Vitvitskyi, Alex, Pascanu, Razvan, Veličković, Petar
We study how information propagates in decoder-only Transformers, which are the architectural backbone of most existing frontier large language models (LLMs). We rely on a theoretical signal propagation analysis -- specifically, we analyse the repres
Externí odkaz:
http://arxiv.org/abs/2406.04267
Autor:
Sharifzadeh, Sahand, Kaplanis, Christos, Pathak, Shreya, Kumaran, Dharshan, Ilic, Anastasija, Mitrovic, Jovana, Blundell, Charles, Banino, Andrea
The creation of high-quality human-labeled image-caption datasets presents a significant bottleneck in the development of Visual-Language Models (VLMs). In this work, we investigate an approach that leverages the strengths of Large Language Models (L
Externí odkaz:
http://arxiv.org/abs/2403.07750
Autor:
Chan, Stephanie C. Y., Dasgupta, Ishita, Kim, Junkyung, Kumaran, Dharshan, Lampinen, Andrew K., Hill, Felix
Transformer models can use two fundamentally different kinds of information: information stored in weights during training, and information provided ``in-context'' at inference time. In this work, we show that transformers exhibit different inductive
Externí odkaz:
http://arxiv.org/abs/2210.05675
Autor:
Dasgupta, Ishita, Lampinen, Andrew K., Chan, Stephanie C. Y., Sheahan, Hannah R., Creswell, Antonia, Kumaran, Dharshan, McClelland, James L., Hill, Felix
Reasoning is a key ability for an intelligent system. Large language models (LMs) achieve above-chance performance on abstract reasoning tasks, but exhibit many imperfections. However, human abstract reasoning is also imperfect. For example, human re
Externí odkaz:
http://arxiv.org/abs/2207.07051
Autor:
Banino, Andrea, Badia, Adrià Puigdomènech, Köster, Raphael, Chadwick, Martin J., Zambaldi, Vinicius, Hassabis, Demis, Barry, Caswell, Botvinick, Matthew, Kumaran, Dharshan, Blundell, Charles
Recent research developing neural network architectures with external memory have often used the benchmark bAbI question and answering dataset which provides a challenging number of tasks requiring reasoning. Here we employed a classic associative in
Externí odkaz:
http://arxiv.org/abs/2001.10913
Autor:
Silver, David, Hubert, Thomas, Schrittwieser, Julian, Antonoglou, Ioannis, Lai, Matthew, Guez, Arthur, Lanctot, Marc, Sifre, Laurent, Kumaran, Dharshan, Graepel, Thore, Lillicrap, Timothy, Simonyan, Karen, Hassabis, Demis
The game of chess is the most widely-studied domain in the history of artificial intelligence. The strongest programs are based on a combination of sophisticated search techniques, domain-specific adaptations, and handcrafted evaluation functions tha
Externí odkaz:
http://arxiv.org/abs/1712.01815
Autor:
Kirkpatrick, James, Pascanu, Razvan, Rabinowitz, Neil, Veness, Joel, Desjardins, Guillaume, Rusu, Andrei A., Milan, Kieran, Quan, John, Ramalho, Tiago, Grabska-Barwinska, Agnieszka, Hassabis, Demis, Clopath, Claudia, Kumaran, Dharshan, Hadsell, Raia
The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of
Externí odkaz:
http://arxiv.org/abs/1612.00796
Autor:
Wang, Jane X, Kurth-Nelson, Zeb, Tirumala, Dhruva, Soyer, Hubert, Leibo, Joel Z, Munos, Remi, Blundell, Charles, Kumaran, Dharshan, Botvinick, Matt
In recent years deep reinforcement learning (RL) systems have attained superhuman performance in a number of challenging task domains. However, a major limitation of such applications is their demand for massive amounts of training data. A critical p
Externí odkaz:
http://arxiv.org/abs/1611.05763
Autor:
Mirowski, Piotr, Pascanu, Razvan, Viola, Fabio, Soyer, Hubert, Ballard, Andrew J., Banino, Andrea, Denil, Misha, Goroshin, Ross, Sifre, Laurent, Kavukcuoglu, Koray, Kumaran, Dharshan, Hadsell, Raia
Learning to navigate in complex environments with dynamic elements is an important milestone in developing AI agents. In this work we formulate the navigation question as a reinforcement learning problem and show that data efficiency and task perform
Externí odkaz:
http://arxiv.org/abs/1611.03673