4 citations · 11 across the 15 of their papers we have counts for
15 papers
Mechanisms of Misgeneralization in Physical Sequence Modeling
Kento Nishi, Raphael Tang, Karun Kumar +2
Generative sequence models are often trained to plan motion in physical domains, from robotics to mechanical simulations. When constructing a dataset to train such a model, enginee…
In-Context Learning Strategies Emerge Rationally
Daniel Wurgaft, Ekdeep Singh Lubana, Core Francisco Park +3
Recent work analyzing in-context learning (ICL) has identified a broad set of strategies that describe model behavior in different experimental conditions. We aim to unify these fi…
Decomposing Elements of Problem Solving: What "Math" Does RL Teach?
Tian Qin, Core Francisco Park, Mujin Kwun +5
Mathematical reasoning tasks have become prominent benchmarks for assessing the reasoning capabilities of LLMs, especially with reinforcement learning (RL) methods such as GRPO sho…
: System-2 Fine-tuning for Robust Integration of New Knowledge
Core Francisco Park, Zechen Zhang, Hidenori Tanaka
Humans and intelligent animals can internalize new information and accurately internalize their implications to perform downstream tasks. While large language models (LLMs) can ach…
ICLR: In-Context Learning of Representations
Core Francisco Park, Andrew Lee, Ekdeep Singh Lubana +5
Recent work has demonstrated that semantics specified by pretraining data influence how representations of different concepts are organized in a large language model (LLM). However…
Competition Dynamics Shape Algorithmic Phases of In-Context Learning
Core Francisco Park, Ekdeep Singh Lubana, Itamar Pres +1
In-Context Learning (ICL) has significantly expanded the general-purpose nature of large language models, allowing them to adapt to novel tasks using merely the inputted context. T…
Representation Shattering in Transformers: A Synthetic Study with Knowledge Editing
Kento Nishi, Rahul Ramesh, Maya Okawa +3
Knowledge Editing (KE) algorithms alter models' weights to perform targeted updates to incorrect, outdated, or otherwise unwanted factual associations. However, recent work has sho…
Swing-by Dynamics in Concept Learning and Compositional Generalization
Yongyi Yang, Core Francisco Park, Ekdeep Singh Lubana +3
Prior work has shown that text-conditioned diffusion models can learn to identify and manipulate primitive concepts underlying a compositional data-generating process, enabling gen…
A Percolation Model of Emergence: Analyzing Transformers Trained on a Formal Language
Ekdeep Singh Lubana, Kyogo Kawaguchi, Robert P. Dick +1
Increase in data, size, or compute can lead to sudden learning of specific capabilities by a neural network -- a phenomenon often called "emergence''. Beyond scientific understandi…
Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space
Core Francisco Park, Maya Okawa, Andrew Lee +2
Modern generative models demonstrate impressive capabilities, likely stemming from an ability to identify and manipulate abstract concepts underlying their training data. However,…
Towards an Understanding of Stepwise Inference in Transformers: A Synthetic Graph Navigation Model
Mikail Khona, Maya Okawa, Jan Hula +5
Stepwise inference protocols, such as scratchpads and chain-of-thought, help language models solve complex problems by decomposing them into a sequence of simpler subproblems. Desp…
Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks
Rahul Ramesh, Ekdeep Singh Lubana, Mikail Khona +2
Transformers trained on huge text corpora exhibit a remarkable set of capabilities, e.g., performing basic arithmetic. Given the inherent compositional nature of language, one can…
Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks
Samyak Jain, Robert Kirk, Ekdeep Singh Lubana +5
Fine-tuning large pre-trained models has become the de facto strategy for developing both task-specific and general-purpose machine learning systems, including developing models th…
In-Context Learning Dynamics with Random Binary Sequences
Eric J. Bigelow, Ekdeep Singh Lubana, Robert P. Dick +2
Large language models (LLMs) trained on huge corpora of text datasets demonstrate intriguing capabilities, achieving state-of-the-art performance on tasks they were not explicitly…
Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task
Maya Okawa, Ekdeep Singh Lubana, Robert P. Dick +1
Modern generative models exhibit unprecedented capabilities to generate extremely realistic data. However, given the inherent compositionality of the real world, reliable use of th…