Tolga Dimlioglu

Tolga Dimlioglu, Kristi Topollai, Anna Choromanska

Deep neural network training often exhibits highly anisotropic loss geometry, where a few sharp dominant Hessian directions coexist with a large flatter bulk. Gradients tend to align disproportionately with these dominant directions, although stable progress often requires movement through flatter bulk directions. Estimating the dominant subspace is therefore useful but costly with direct Hessian-based methods. We show that standard Local SGD exposes this geometry through worker disagreement. We theoretically show that the worker-average gap covariance is shaped by stochastic-gradient noise and Hessian curvature, causing workers to disagree along sharp, curvature-sensitive directions. Thus, worker-average gaps provide a cheap Hessian-free estimator of the dominant subspace. Experiments on MLPs, CNNs, and Transformers show that subspaces formed by worker-average gaps capture a substantial fraction of the gradient component lying in the dominant Hessian eigenspace.

Kristi Topollai, Allan Ma, Tolga Dimlioglu et al.

Communication-efficient distributed optimizers such as DiLoCo reduce synchronization costs by letting workers perform many local updates before aggregating their progress with an outer momentum optimizer. Recent theory suggests that the outer optimizer acts on an effective spectrum induced by the inner optimization loop, and that the choice of outer momentum controls how progress from local updates is accumulated across communication rounds. We study periodic restarting of the outer momentum as a simple complementary mechanism for controlling this outer memory. In a linearized squared-loss model where prediction-space residuals evolve under the empirical NTK, we derive a mode-wise restart contraction showing that resets exploit phase cancellation by discarding stale momentum while preserving inner-loop progress. Toy experiments verify the predicted contraction behavior, and language-model pretraining experiments show that periodic restarts widen the stable range of outer learning rates and momentum values across communication periods.

Tolga Dimlioglu, Anna Choromanska

We study distributed training of deep learning models in time-constrained environments. We propose a new algorithm that periodically pulls workers towards the center variable computed as a weighted average of workers, where the weights are inversely proportional to the gradient norms of the workers such that recovering the flat regions in the optimization landscape is prioritized. We develop two asynchronous variants of the proposed algorithm that we call Model-level and Layer-level Gradient-based Weighted Averaging (resp. MGRAWA and LGRAWA), which differ in terms of the weighting scheme that is either done with respect to the entire model or is applied layer-wise. On the theoretical front, we prove the convergence guarantee for the proposed approach in both convex and non-convex settings. We then experimentally demonstrate that our algorithms outperform the competitor methods by achieving faster convergence and recovering better quality and flatter local optima. We also carry out an ablation study to analyze the scalability of the proposed algorithms in more crowded distributed training environments. Finally, we report that our approach requires less frequent communication and fewer distributed updates compared to the state-of-the-art baselines.

Alexander Naumann, Xunjiang Gu, Tolga Dimlioglu et al.

Autonomous vehicle (AV) stacks have traditionally relied on decomposed approaches, with separate modules handling perception, prediction, and planning. However, this design introduces information loss during inter-module communication, increases computational overhead, and can lead to compounding errors. To address these challenges, recent works have proposed architectures that integrate all components into an end-to-end differentiable model, enabling holistic system optimization. This shift emphasizes data engineering over software integration, offering the potential to enhance system performance by simply scaling up training resources. In this work, we evaluate the performance of a simple end-to-end driving architecture on internal driving datasets ranging in size from 16 to 8192 hours with both open-loop metrics and closed-loop simulations. Specifically, we investigate how much additional training data is needed to achieve a target performance gain, e.g., a 5% improvement in motion prediction accuracy. By understanding the relationship between model performance and training dataset size, we aim to provide insights for data-driven decision-making in autonomous driving development.

Tolga Dimlioglu, Nadine Chang, Maying Shen et al.

Large-scale deep learning models for physical AI applications depend on diverse training data collection efforts. These models and correspondingly, the training data, must address different evaluation criteria necessary for the models to be deployable in real-world environments. Data selection policies can guide the development of the training set, but current frameworks do not account for the ambiguity in how data points affect different metrics. In this work, we propose Mixture Optimization via Scaling-Aware Iterative Collection (MOSAIC), a general data selection framework that operates by: (i) partitioning the dataset into domains; (ii) fitting neural scaling laws from each data domain to the evaluation metrics; and (iii) optimizing a data mixture by iteratively adding data from domains that maximize the change in metrics. We apply MOSAIC to autonomous driving (AD), where an End-to-End (E2E) planner model is evaluated on the Extended Predictive Driver Model Score (EPDMS), an aggregate of driving rule compliance metrics. Here, MOSAIC outperforms a diverse set of baselines on EPDMS with up to 80\% less data.

Kristi Topollai, Tolga Dimlioglu, Anna Choromanska et al.

Contract management involves reviewing and negotiating provisions, individual clauses that define rights, obligations, and terms of agreement. During this process, revisions to provisions are proposed and iteratively refined, some of which may be problematic or unacceptable. Automating this workflow is challenging due to the scarcity of labeled data and the abundance of unstructured legacy contracts. In this paper, we present a modular framework designed to streamline contract management through a retrieval-augmented generation (RAG) pipeline. Our system integrates synthetic data generation, semantic clause retrieval, acceptability classification, and reward-based alignment to flag problematic revisions and generate improved alternatives. Developed and evaluated in collaboration with an industry partner, our system achieves over 80% accuracy in both identifying and optimizing problematic revisions, demonstrating strong performance under real-world, low-resource conditions and offering a practical means of accelerating contract revision workflows.

Tolga Dimlioglu, Anna Choromanska

We study centralized distributed data parallel training of deep neural networks (DNNs), aiming to improve the trade-off between communication efficiency and model performance of the local gradient methods. To this end, we revisit the flat-minima hypothesis, which suggests that models with better generalization tend to lie in flatter regions of the loss landscape. We introduce a simple, yet effective, sharpness measure, Inverse Mean Valley, and demonstrate its strong correlation with the generalization gap of DNNs. We incorporate an efficient relaxation of this measure into the distributed training objective as a lightweight regularizer that encourages workers to collaboratively seek wide minima. The regularizer exerts a pushing force that counteracts the consensus step pulling the workers together, giving rise to the Distributed Pull-Push Force (DPPF) algorithm. Empirically, we show that DPPF outperforms other communication-efficient approaches and achieves better generalization performance than local gradient methods and synchronous gradient averaging, while maintaining communication efficiency. In addition, our loss landscape visualizations confirm the ability of DPPF to locate flatter minima. On the theoretical side, we show that DPPF guides workers to span flat valleys, with the final valley width governed by the interplay between push and pull strengths, and that its pull-push dynamics is self-stabilizing. We further provide generalization guarantees linked to the valley width and prove convergence in the non-convex setting.

Nikhil Churamani, Tolga Dimlioglu, German I. Parisi et al.

Understanding human affective behaviour, especially in the dynamics of real-world settings, requires Facial Expression Recognition (FER) models to continuously adapt to individual differences in user expression, contextual attributions, and the environment. Current (deep) Machine Learning (ML)-based FER approaches pre-trained in isolation on benchmark datasets fail to capture the nuances of real-world interactions where data is available only incrementally, acquired by the agent or robot during interactions. New learning comes at the cost of previous knowledge, resulting in catastrophic forgetting. Lifelong or Continual Learning (CL), on the other hand, enables adaptability in agents by being sensitive to changing data distributions, integrating new information without interfering with previously learnt knowledge. Positing CL as an effective learning paradigm for FER, this work presents the Continual Facial Expression Recognition (ConFER) benchmark that evaluates popular CL techniques on FER tasks. It presents a comparative analysis of several CL-based approaches on popular FER datasets such as CK+, RAF-DB, and AffectNet and present strategies for a successful implementation of ConFER for Affective Computing (AC) research. CL techniques, under different learning settings, are shown to achieve state-of-the-art (SOTA) performance across several datasets, thus motivating a discussion on the benefits of applying CL principles towards human behaviour understanding, particularly from facial expressions, as well the challenges entailed.