Author name cluster

Ege Beyazit

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

2 author rows

TMLR Journal 2026 Journal Article

$$\texttt{C2-DPO}$$: Constrained Controlled Direct Preference Optimization

Kavosh Asadi
Xingzi Xu
Julien Han
Ege Beyazit
Idan Pipano
Dominique Perrault-Joncas
Shoham Sabach
Mohammad Ghavamzadeh

Direct preference optimization (\texttt{DPO}) has emerged as a promising approach for solving the alignment problem in AI. In this paper, we make two counter-intuitive observations about \texttt{DPO}. First, we show that the \texttt{DPO} loss could be derived by starting from an alternative optimization problem that only defines the KL guardrail on in-sample responses, unlike the original RLHF problem where guardrails are defined on the entire distribution. Second, we prove a surprising property of this alternative optimization problem, where both the preferred and rejected responses tend to decrease in probability under its optimal policy, a phenomenon typically displayed by DPO in practice. To control this behavior, we propose a set of constraints designed to limit the displacement of probability mass between the preferred and rejected responses in the reference and target policies. The resulting algorithm, which we call Constrained Controlled DPO (\texttt{C2-DPO}), has a meaningful RLHF interpretation. By hedging against the displacement, \texttt{C2-DPO} provides practical improvements over vanilla \texttt{DPO} when aligning several language models using standard preference datasets.

PDF Details

NeurIPS Conference 2023 Conference Paper

An Inductive Bias for Tabular Deep Learning

Ege Beyazit
Jonathan Kozaczuk
Bo Li
Vanessa Wallace
Bilal Fadlallah

Deep learning methods have achieved state-of-the-art performance in most modeling tasks involving images, text and audio, however, they typically underperform tree-based methods on tabular data. In this paper, we hypothesize that a significant contributor to this performance gap is the interaction between irregular target functions resulting from the heterogeneous nature of tabular feature spaces, and the well-known tendency of neural networks to learn smooth functions. Utilizing tools from spectral analysis, we show that functions described by tabular datasets often have high irregularity, and that they can be smoothed by transformations such as scaling and ranking in order to improve performance. However, because these transformations tend to lose information or negatively impact the loss landscape during optimization, they need to be rigorously fine-tuned for each feature to achieve performance gains. To address these problems, we propose introducing frequency reduction as an inductive bias. We realize this bias as a neural network layer that promotes learning low-frequency representations of the input features, allowing the network to operate in a space where the target function is more regular. Our proposed method introduces less computational complexity than a fully connected layer, while significantly improving neural network performance, and speeding up its convergence on 14 tabular datasets.

PDF Details

ECAI Conference 2020 Conference Paper

Online Learning to Accelerate Neural Network Inference with Traveling Classifiers

Ege Beyazit
Yi He 0007
Nian-Feng Tzeng
Xindong Wu 0001

Deep neural networks trained on millions of instances can recognize a wide variety of patterns. It is common to use these pre-trained deep networks in applications where the domain specific training data is not readily available. Once a pre-trained network is deployed to such applications, some of the information contained in the network may be irrelevant due to the difference between the training set and the application data distributions. As a result, parts of the neural network become redundant and slow down inference. This redundancy is unknown until the model is deployed and input data is received. Therefore, it can only be identified and avoided in real-time. Existing works on neural network acceleration can not exploit such redundancy during offline training when the domain-specific datasets are unavailable. In this paper, we study online learning to accelerate neural network inference. We propose traveling classifiers that continuously learn from the activations of two consecutive network layers to accelerate inference in real-time. Traveling classifiers model class conditional probabilities to generate early predictions and bypass unnecessary computation of network layers. The classifiers also adaptively switch the layers they learn from by measuring the feature space differences between the activations. This traveling mechanism automatically adjusts the aggressiveness of the acceleration without sacrificing prediction accuracy. We demonstrate the performance of the proposed algorithm on the ImageNet dataset [10] using the state-of-the-art ResNet-50, ResNet-152 [18] and VGG-16 [38] architectures. Experiments demonstrate that our method significantly outperforms baseline approaches.

Details

IJCAI Conference 2019 Conference Paper

Online Learning from Capricious Data Streams: A Generative Approach

Yi He
Baijun Wu
Di Wu
Ege Beyazit
Sheng Chen
Xindong Wu

Learning with streaming data has received extensive attention during the past few years. Existing approaches assume the feature space is fixed or changes by following explicit regularities, limiting their applicability in dynamic environments where the data streams are described by an arbitrarily varying feature space. To handle such capricious data streams, we in this paper develop a novel algorithm, named OCDS (Online learning from Capricious Data Streams), which does not make any assumption on feature space dynamics. OCDS trains a learner on a universal feature space that establishes relationships between old and new features, so that the patterns learned in the old feature space can be used in the new feature space. Specifically, the universal feature space is constructed by leveraging the relatednesses among features. We propose a generative graphical model to model the construction process, and show that learning from the universal feature space can effectively improve performance with theoretical analysis. The experimental results demonstrate that OCDS achieves conspicuous performance on synthetic and real datasets.

PDF Details

AAAI Conference 2019 Conference Paper

Online Learning from Data Streams with Varying Feature Spaces

Ege Beyazit
Jeevithan Alagurajah
Xindong Wu

We study the problem of online learning with varying feature spaces. The problem is challenging because, unlike traditional online learning problems, varying feature spaces can introduce new features or stop having some features without following a pattern. Other existing methods such as online streaming feature selection (Wu et al. 2013), online learning from trapezoidal data streams (Zhang et al. 2016), and learning with feature evolvable streams (Hou, Zhang, and Zhou 2017) are not capable to learn from arbitrarily varying feature spaces because they make assumptions about the feature space dynamics. In this paper, we propose a novel online learning algorithm OLVF to learn from data with arbitrarily varying feature spaces. The OLVF algorithm learns to classify the feature spaces and the instances from feature spaces simultaneously. To classify an instance, the algorithm dynamically projects the instance classifier and the training instance onto their shared feature subspace. The feature space classifier predicts the projection confidences for a given feature space. The instance classifier will be updated by following the empirical risk minimization principle and the strength of the constraints will be scaled by the projection confidences. Afterwards, a feature sparsity method is applied to reduce the model complexity. Experiments on 10 datasets with varying feature spaces have been conducted to demonstrate the performance of the proposed OLVF algorithm. Moreover, experiments with trapezoidal data streams on the same datasets have been conducted to show that OLVF performs better than the state-of-the-art learning algorithm (Zhang et al. 2016).

PDF Details