13 citations · 22 across the 15 of their papers we have counts for
16 papers
Understanding Contextual Recall in Transformers: How Finetuning Enables In-Context Reasoning over Pretraining Knowledge
Bhavya Vasudeva, Puneesh Deora, Alberto Bietti +2
Transformer-based language models excel at in-context learning (ICL), where they can adapt to new tasks based on contextual examples, without parameter updates. In a specific form…
Why Loss Re-weighting Works If You Stop Early: Training Dynamics of Unconstrained Features
Yize Zhao, Christos Thrampoulidis
The application of loss reweighting in modern deep learning presents a nuanced picture. While it fails to alter the terminal learning phase in overparameterized deep neural network…
Facts in Stats: Impacts of Pretraining Diversity on Language Model Generalization
Tina Behnia, Puneesh Deora, Christos Thrampoulidis
Language models are pretrained on sequences that blend statistical regularities (making text fluent) with factual associations between specific tokens (knowledge of facts). While r…
How Muon's Spectral Design Benefits Generalization: A Study on Imbalanced Data
Bhavya Vasudeva, Puneesh Deora, Yize Zhao +2
The growing adoption of spectrum-aware matrix-valued optimizers such as Muon and Shampoo in deep learning motivates a systematic study of their generalization properties and, in pa…
In-Context Occam's Razor: How Transformers Prefer Simpler Hypotheses on the Fly
Puneesh Deora, Bhavya Vasudeva, Tina Behnia +1
In-context learning (ICL) enables transformers to adapt to new tasks through contextual examples without parameter updates. While existing research has typically studied ICL in fix…
Geometry of Semantics in Next-Token Prediction: How Optimization Implicitly Organizes Linguistic Representations
Yize Zhao, Christos Thrampoulidis
We investigate how next-token prediction (NTP) optimization leads language models to extract and organize semantic structure from text. Our analysis, based on a tractable mathemati…
Implicit Geometry of Next-token Prediction: From Language Sparsity Patterns to Model Representations
Yize Zhao, Tina Behnia, Vala Vakilian +1
Next-token prediction (NTP) over large text corpora has become the go-to paradigm to train large language models. Yet, it remains unclear how NTP influences the mapping of linguist…
Supervised Contrastive Representation Learning: Landscape Analysis with Unconstrained Features
Tina Behnia, Christos Thrampoulidis
Recent findings reveal that over-parameterized deep neural networks, trained beyond zero training-error, exhibit a distinctive structural pattern at the final layer, termed as Neur…
Implicit Optimization Bias of Next-Token Prediction in Linear Models
Christos Thrampoulidis
We initiate an investigation into the optimization properties of next-token prediction (NTP), the dominant training paradigm for modern language models. Specifically, we study the…
Implicit Bias and Fast Convergence Rates for Self-attention
Bhavya Vasudeva, Puneesh Deora, Christos Thrampoulidis
We study the fundamental optimization principles of self-attention, the defining mechanism of transformers, by analyzing the implicit bias of gradient-based optimizers in training…
Engineering the Neural Collapse Geometry of Supervised-Contrastive Loss
Jaidev Gill, Vala Vakilian, Christos Thrampoulidis
Supervised-contrastive loss (SCL) is an alternative to cross-entropy (CE) for classification tasks that makes use of similarities in the embedding space to allow for richer represe…
On the Optimization and Generalization of Multi-head Attention
Puneesh Deora, Rouzbeh Ghaderi, Hossein Taheri +1
The training and generalization dynamics of the Transformer's core mechanism, namely the Attention mechanism, remain under-explored. Besides, existing analyses primarily focus on s…
Symmetric Neural-Collapse Representations with Supervised Contrastive Loss: The Impact of ReLU and Batching
Ganesh Ramachandra Kini, Vala Vakilian, Tina Behnia +2
Supervised contrastive loss (SCL) is a competitive and often superior alternative to the cross-entropy loss for classification. While prior studies have demonstrated that both loss…
On the Implicit Geometry of Cross-Entropy Parameterizations for Label-Imbalanced Data
Tina Behnia, Ganesh Ramachandra Kini, Vala Vakilian +1
Various logit-adjusted parameterizations of the cross-entropy (CE) loss have been proposed as alternatives to weighted CE for training large models on label-imbalanced data far bey…
Imbalance Trouble: Revisiting Neural-Collapse Geometry
Christos Thrampoulidis, Ganesh R. Kini, Vala Vakilian +1
Neural Collapse refers to the remarkable structural properties characterizing the geometry of class embeddings and classifier weights, found by deep nets when trained beyond zero t…
On how to avoid exacerbating spurious correlations when models are overparameterized
Tina Behnia, Ke Wang, Christos Thrampoulidis
Overparameterized models fail to generalize well in the presence of data imbalance even when combined with traditional techniques for mitigating imbalances. This paper focuses on i…