Zafer Doğan

Eser Ilke Genc, Samet Demir, Zafer Dogan

Independent Component Analysis (ICA) is a foundational tool for unsupervised representation learning, yet its high-dimensional theory remains largely limited to single-component recovery. We develop an asymptotically exact mean-field theory for multi-component online ICA, capturing the coupling induced by simultaneous learning and orthogonalization. In the high-dimensional limit, the joint empirical distribution of learned estimates and ground-truth components converges to a deterministic process, yielding a closed ODE system for the overlap matrix between learned directions and true components. This characterization reveals a genuinely multi-component, initialization-driven phase structure: a decoupled regime, where estimates align with distinct components and evolve nearly independently, and a competition regime, where overlapping initializations induce orthogonality-driven conflicts, slow reorientation, and delayed convergence. Our steady-state analysis gives explicit learnability boundaries and competition conditions linking step size, data moments, and initialization. These conditions show that larger higher-order moments and competition shrink the stable learning-rate window, increase convergence times, and predict a staircase phenomenon in which the number of recoverable components changes discretely with the learning rate. Experiments on synthetic data and hyperspectral remote sensing data validate the predicted trajectories and phase behavior.

Samet Demir, Zafer Dogan

Random feature models (RFMs), two-layer networks with a randomly initialized fixed first layer and a trained linear readout, are among the simplest nonlinear predictors. Prior asymptotic analyses in the proportional high-dimensional regime show that, under isotropic data, RFMs reduce to noisy linear models and offer no advantage over classical linear methods such as ridge regression. Yet RFMs frequently outperform linear baselines on structured real data. We show that this tension is explained by a correlation-driven phase transition: under spiked-covariance designs, the interaction between anisotropy and input-label correlation determines whether the RFM behaves as an effectively linear predictor or exhibits genuinely nonlinear gains. Concretely, we establish a universality principle under anisotropy and characterize the RFM generalization error via an equivalent noisy polynomial model. The effective degree of this polynomial, equivalently, which Hermite orders of the activation survive, is governed by the strength of input-label correlation, yielding an explicit boundary in the correlation-spike-magnitude plane. Below the boundary, the RFM collapses to a linear surrogate and can underperform strong linear baselines; above it, higher-order terms persist and the RFM achieves a clear nonlinear advantage. Numerical simulations and real-data experiments corroborate the theory and delineate the transition between these two regimes.

Samet Demir, Zafer Doğan

Random feature model with a nonlinear activation function has been shown to perform asymptotically equivalent to a Gaussian model in terms of training and generalization errors. Analysis of the equivalent model reveals an important yet not fully understood role played by the activation function. To address this issue, we study the "parameters" of the equivalent model to achieve improved generalization performance for a given supervised learning problem. We show that acquired parameters from the Gaussian model enable us to define a set of optimal nonlinearities. We provide two example classes from this set, e.g., second-order polynomial and piecewise linear functions. These functions are optimized to improve generalization performance regardless of the actual form. We experiment with regression and classification problems, including synthetic and real (e.g., CIFAR10) data. Our numerical results validate that the optimized nonlinearities achieve better generalization performance than widely-used nonlinear functions such as ReLU. Furthermore, we illustrate that the proposed nonlinearities also mitigate the so-called double descent phenomenon, which is known as the non-monotonic generalization performance regarding the sample size and the model size.

Samet Demir, Zafer Dogan

Many online learning algorithms, including classical online PCA methods, enforce explicit normalization steps that discard the evolving norm of the parameter vector. We show that this norm can in fact encode meaningful information about the underlying statistical structure of the problem, and that exploiting this information leads to improved learning behavior. Motivated by this principle, we introduce Implicitly Normalized Online PCA (INO-PCA), an online PCA algorithm that removes the unit-norm constraint and instead allows the parameter norm to evolve dynamically through a simple regularized update. We prove that in the high-dimensional limit the joint empirical distribution of the estimate and the true component converges to a deterministic measure-valued process governed by a nonlinear PDE. This analysis reveals that the parameter norm obeys a closed-form ODE coupled with the cosine similarity, forming an internal state variable that regulates learning rate, stability, and sensitivity to signal-to-noise ratio (SNR). The resulting dynamics uncover a three-way relationship between the norm, SNR, and optimal step size, and expose a sharp phase transition in steady-state performance. Both theoretically and experimentally, we show that INO-PCA consistently outperforms Oja's algorithm and adapts rapidly in non-stationary environments. Overall, our results demonstrate that relaxing norm constraints can be a principled and effective way to encode and exploit problem-relevant information in online learning algorithms.

Onat Ure, Samet Demir, Zafer Dogan

We study the effect of high-order statistics of data on the learning dynamics of neural networks (NNs) by using a moment-controllable non-Gaussian data model. Considering the expressivity of two-layer neural networks, we first construct the data model as a generative two-layer NN where the activation function is expanded by using Hermite polynomials. This allows us to achieve interpretable control over high-order cumulants such as skewness and kurtosis through the Hermite coefficients while keeping the data model realistic. Using samples generated from the data model, we perform controlled online learning experiments with a two-layer NN. Our results reveal a moment-wise progression in training: networks first capture low-order statistics such as mean and covariance, and progressively learn high-order cumulants. Finally, we pretrain the generative model on the Fashion-MNIST dataset and leverage the generated samples for further experiments. The results of these additional experiments confirm our conclusions and show the utility of the data model in a real-world scenario. Overall, our proposed approach bridges simplified data assumptions and practical data complexity, which offers a principled framework for investigating distributional effects in machine learning and signal processing.

Samet Demir, Zafer Dogan

Pretrained Transformers excel at in-context learning (ICL), inferring new tasks from only a handful of examples. Yet, their ICL performance can degrade sharply under distribution shift between pretraining and test data, a regime increasingly common in real-world deployments. While recent empirical work hints that adjusting the attention temperature in the softmax can enhance Transformer performance, the attention temperature's role in ICL under distribution shift remains unexplored. This paper provides the first theoretical and empirical study of attention temperature for ICL under distribution shift. Using a simplified but expressive "linearized softmax" framework, we derive closed-form generalization error expressions and prove that shifts in input covariance or label noise substantially impair ICL, but that an optimal attention temperature exists which minimizes this error. We then validate our predictions through extensive simulations on linear regression tasks and large-scale experiments with GPT-2 and LLaMA2-7B on question-answering benchmarks. Our results establish attention temperature as a principled and powerful mechanism for improving the robustness of ICL in pretrained Transformers, advancing theoretical understanding and providing actionable guidance for selecting attention temperature in practice.

Cansu Korkmaz, A. Murat Tekalp, Zafer Dogan

Super-resolution (SR) is an ill-posed inverse problem, where the size of the set of feasible solutions that are consistent with a given low-resolution image is very large. Many algorithms have been proposed to find a "good" solution among the feasible solutions that strike a balance between fidelity and perceptual quality. Unfortunately, all known methods generate artifacts and hallucinations while trying to reconstruct high-frequency (HF) image details. A fundamental question is: Can a model learn to distinguish genuine image details from artifacts? Although some recent works focused on the differentiation of details and artifacts, this is a very challenging problem and a satisfactory solution is yet to be found. This paper shows that the characterization of genuine HF details versus artifacts can be better learned by training GAN-based SR models using wavelet-domain loss functions compared to RGB-domain or Fourier-space losses. Although wavelet-domain losses have been used in the literature before, they have not been used in the context of the SR task. More specifically, we train the discriminator only on the HF wavelet sub-bands instead of on RGB images and the generator is trained by a fidelity loss over wavelet subbands to make it sensitive to the scale and orientation of structures. Extensive experimental results demonstrate that our model achieves better perception-distortion trade-off according to multiple objective measures and visual evaluations.

Andrew Bond, Zafer Doğan

We study a solvable high-dimensional model of generative adversarial network (GAN) training in which a linear generator learns a low-dimensional subspace from data with structured latent covariance. Prior solvable GAN analyses assume unconditional signals with diagonal latent covariance; we extend the multi-feature discriminator setting to class-dependent, correlated, and non-zero-mean latent structure. For the quadratic energy discriminator, all such heterogeneity enters the dynamics through a probability-weighted effective second moment. We prove that the stochastic microscopic training process converges, in the high-dimensional limit, to deterministic ordinary differential equations governed by this effective covariance. In the matched-covariance specialization, the stability analysis yields a mode-wise solvable interval determined by the learning rates and noise level: learning begins when the leading effective eigenvalue crosses the lower threshold, while full recovery requires all relevant effective modes to remain within the interval. This reveals a signal-boosting mechanism: low-rank correlations can lift weak directions above the learnability threshold, whereas overly strong correlations destabilize recovery. Numerical simulations validate the ODE, phase boundary, and boosting mechanism. Experiments on MNIST, FashionMNIST, and CIFAR-10 further show that informed generator covariance improves alignment with the data-driven reference subspace.

Samet Demir, Zafer Dogan

We study the in-context learning (ICL) capabilities of pretrained Transformers in the setting of nonlinear regression. Specifically, we focus on a random Transformer with a nonlinear MLP head where the first layer is randomly initialized and fixed while the second layer is trained. Furthermore, we consider an asymptotic regime where the context length, input dimension, hidden dimension, number of training tasks, and number of training samples jointly grow. In this setting, we show that the random Transformer behaves equivalent to a finite-degree Hermite polynomial model in terms of ICL error. This equivalence is validated through simulations across varying activation functions, context lengths, hidden layer widths (revealing a double-descent phenomenon), and regularization settings. Our results offer theoretical and empirical insights into when and how MLP layers enhance ICL, and how nonlinearity and over-parameterization influence model performance.

Cansu Korkmaz, Ahmet Murat Tekalp, Zafer Dogan

Super-resolution (SR) is an ill-posed inverse problem with many feasible solutions consistent with a given low-resolution image. On one hand, regressive SR models aim to balance fidelity and perceptual quality to yield a single solution, but this trade-off often introduces artifacts that create ambiguity in information-critical applications such as recognizing digits or letters. On the other hand, diffusion models generate a diverse set of SR images, but selecting the most trustworthy solution from this set remains a challenge. This paper introduces a robust, automated framework for identifying the most trustworthy SR sample from a diffusion-generated set by leveraging the semantic reasoning capabilities of vision-language models (VLMs). Specifically, VLMs such as BLIP-2, GPT-4o, and their variants are prompted with structured queries to assess semantic correctness, visual quality, and artifact presence. The top-ranked SR candidates are then ensembled to yield a single trustworthy output in a cost-effective manner. To rigorously assess the validity of VLM-selected samples, we propose a novel Trustworthiness Score (TWS) a hybrid metric that quantifies SR reliability based on three complementary components: semantic similarity via CLIP embeddings, structural integrity using SSIM on edge maps, and artifact sensitivity through multi-level wavelet decomposition. We empirically show that TWS correlates strongly with human preference in both ambiguous and natural images, and that VLM-guided selections consistently yield high TWS values. Compared to conventional metrics like PSNR, LPIPS, which fail to reflect information fidelity, our approach offers a principled, scalable, and generalizable solution for navigating the uncertainty of the diffusion SR space. By aligning outputs with human expectations and semantic correctness, this work sets a new benchmark for trustworthiness in generative SR.

Cansu Korkmaz, Ege Cirakman, A. Murat Tekalp et al.

Super-resolution (SR) is an ill-posed inverse problem with a large set of feasible solutions that are consistent with a given low-resolution image. Various deterministic algorithms aim to find a single solution that balances fidelity and perceptual quality; however, this trade-off often causes visual artifacts that bring ambiguity in information-centric applications. On the other hand, diffusion models (DMs) excel in generating a diverse set of feasible SR images that span the solution space. The challenge is then how to determine the most likely solution among this set in a trustworthy manner. We observe that quantitative measures, such as PSNR, LPIPS, DISTS, are not reliable indicators to resolve ambiguous cases. To this effect, we propose employing human feedback, where we ask human subjects to select a small number of likely samples and we ensemble the averages of selected samples. This strategy leverages the high-quality image generation capabilities of DMs, while recognizing the importance of obtaining a single trustworthy solution, especially in use cases, such as identification of specific digits or letters, where generating multiple feasible solutions may not lead to a reliable outcome. Experimental results demonstrate that our proposed strategy provides more trustworthy solutions when compared to state-of-the art SR methods.

Andrew Bond, Zafer Dogan

Subspace learning is a critical endeavor in contemporary machine learning, particularly given the vast dimensions of modern datasets. In this study, we delve into the training dynamics of a single-layer GAN model from the perspective of subspace learning, framing these GANs as a novel approach to this fundamental task. Through a rigorous scaling limit analysis, we offer insights into the behavior of this model. Extending beyond prior research that primarily focused on sequential feature learning, we investigate the non-sequential scenario, emphasizing the pivotal role of inter-feature interactions in expediting training and enhancing performance, particularly with an uninformed initialization strategy. Our investigation encompasses both synthetic and real-world datasets, such as MNIST and Olivetti Faces, demonstrating the robustness and applicability of our findings to practical scenarios. By bridging our analysis to the realm of subspace learning, we systematically compare the efficacy of GAN-based methods against conventional approaches, both theoretically and empirically. Notably, our results unveil that while all methodologies successfully capture the underlying subspace, GANs exhibit a remarkable capability to acquire a more informative basis, owing to their intrinsic ability to generate new data samples. This elucidates the unique advantage of GAN-based approaches in subspace learning tasks.

Yigit E. Yildirim, Samet Demir, Zafer Dogan

Empirical Risk Minimization (ERM) is a foundational framework for supervised learning but primarily optimizes average-case performance, often neglecting fairness and robustness considerations. Tilted Empirical Risk Minimization (TERM) extends ERM by introducing an exponential tilt hyperparameter $t$ to balance average-case accuracy with worst-case fairness and robustness. However, in online or streaming settings where data arrive one sample at a time, the classical TERM objective degenerates to standard ERM, losing tilt sensitivity. We address this limitation by proposing an online TERM formulation that removes the logarithm from the classical objective, preserving tilt effects without additional computational or memory overhead. This formulation enables a continuous trade-off controlled by $t$, smoothly interpolating between ERM ($t \to 0$), fairness emphasis ($t > 0$), and robustness to outliers ($t < 0$). We empirically validate online TERM on two representative streaming tasks: robust linear regression with adversarial outliers and minority-class detection in binary classification. Our results demonstrate that negative tilting effectively suppresses outlier influence, while positive tilting improves recall with minimal impact on precision, all at per-sample computational cost equivalent to ERM. Online TERM thus recovers the full robustness-fairness spectrum of classical TERM in an efficient single-sample learning regime.

M. Oguzhan Gultekin, Samet Demir, Zafer Dogan

We investigate the impact of high-order moments on the learning dynamics of an online Independent Component Analysis (ICA) algorithm under a high-dimensional data model composed of a weighted sum of two non-Gaussian random variables. This model allows precise control of the input moment structure via a weighting parameter. Building on an existing ordinary differential equation (ODE)-based analysis in the high-dimensional limit, we demonstrate that as the high-order moments increase, the algorithm exhibits slower convergence and demands both a lower learning rate and greater initial alignment to achieve informative solutions. Our findings highlight the algorithm's sensitivity to the statistical structure of the input data, particularly its moment characteristics. Furthermore, the ODE framework reveals a critical learning rate threshold necessary for learning when moments approach their maximum. These insights motivate future directions in moment-aware initialization and adaptive learning rate strategies to counteract the degradation in learning speed caused by high non-Gaussianity, thereby enhancing the robustness and efficiency of ICA in complex, high-dimensional settings.

Onur Keleş, M. Akın Yılmaz, A. Murat Tekalp et al.

When comparing learned image/video restoration and compression methods, it is common to report peak-signal to noise ratio (PSNR) results. However, there does not exist a generally agreed upon practice to compute PSNR for sets of images or video. Some authors report average of individual image/frame PSNR, which is equivalent to computing a single PSNR from the geometric mean of individual image/frame mean-square error (MSE). Others compute a single PSNR from the arithmetic mean of frame MSEs for each video. Furthermore, some compute the MSE/PSNR of Y-channel only, while others compute MSE/PSNR for RGB channels. This paper investigates different approaches to computing PSNR for sets of images, single video, and sets of video and the relation between them. We show the difference between computing the PSNR based on arithmetic vs. geometric mean of MSE depends on the distribution of MSE over the set of images or video, and that this distribution is task-dependent. In particular, these two methods yield larger differences in restoration problems, where the MSE is exponentially distributed and smaller differences in compression problems, where the MSE distribution is narrower. We hope this paper will motivate the community to clearly describe how they compute reported PSNR values to enable consistent comparison.