Analog in-memory computing (AIMC) accelerators can deliver significant energy efficiency over conventional architectures, but their accuracy is limited by device and circuit-level non-idealities. While prior work has characterized the effects of these non-idealities at model or layer granularity, their impact on individual samples remains largely unexplored. In this work, we show that an input’s vulnerability to such non-idealities can be strongly predicted by its loss curvature, a metric capturing how sharply the loss changes under small input perturbations. Across multiple models, datasets, and non-idealities, our experiments reveal a strong positive correlation between input loss curvature and analog non-ideality-induced failures, with failure rates increasing substantially for high-curvature samples. These findings uncover a previously overlooked, sample-level dimension of hardware robustness and suggest new opportunities for input-aware strategies for addressing non-idealities.

Label corruption, where training samples are mislabeled due to non-expert annotation or adversarial attacks, significantly degrades model performance. Acquiring large, perfectly labeled datasets is costly, and retraining models from scratch is computationally expensive. To address this, we introduce Scaled Activation Projection (SAP), a novel SVD (Singular Value Decomposition)-based corrective machine unlearning algorithm. SAP mitigates label noise by identifying a small subset of trusted samples using cross-entropy loss and projecting model weights onto a clean activation space estimated using SVD on these trusted samples. This process suppresses the noise introduced in activations due to the mislabeled samples. In our experiments, we demonstrate SAP’s effectiveness on synthetic noise with different settings and real-world label noise. SAP applied to the CIFAR dataset with 25% synthetic corruption show upto 6% generalization improvements. Additionally, SAP can improve the generalization over noise robust training approaches on CIFAR dataset by ∼ 3.2% on average. Further, we observe generalization improvements of 2.31% for a Vision Transformer model trained on naturally corrupted Clothing1M.

Deep learning has become the de facto approach in nearly all learning tasks. It has been observed that deep models tend to memorize and sometimes overfit data, which can lead to compromises in performance, privacy, and other critical metrics. In this paper, we explore the theoretical foundations that connect memorization to sample loss, focusing on learning dynamics to understand what and how deep models memorize. To this end, we introduce a novel proxy for memorization: Cumulative Sample Loss (CSL). CSL represents the accumulated loss of a sample throughout the training process. CSL exhibits remarkable similarity to stability-based memorization, as evidenced by considerably high cosine similarity scores. We delve into the theory behind these results, demonstrating that low CSL leads to nontrivial bounds on the extent of stability-based memorization and learning time. The proposed proxy, CSL, is four orders of magnitude less computationally expensive than the stability-based method and can be obtained with zero additional overhead during training. We demonstrate the practical utility of the proposed proxy in identifying mislabeled samples and detecting duplicates where our metric achieves state-of-the-art performance.

Data quality is fundamental to machine-learning performance. An understanding of the data distribution directly influences model security and robustness. By analyzing the curvature of the loss landscape with respect to data, we can detect regions of the distribution that are well represented, and samples which are under-represented. Samples in these under-represented “tails” are disproportionately sensitive to hardware and adversarial noise and are therefore less robust. Compounding the problem, models tend to memorize these rare examples, increasing privacy risk. Curvature analysis helps expose such vulnerabilities by powering stronger membership-inference attacks. Synthetic-data techniques such as DP-ImySyn [24] can be used to augment scarce data, strengthening coverage. Ultimately, building resilient, robust, and secure models begins with understanding the data and fortifying its weak points.

Deep neural networks are over-parameterized and easily overfit to and memorize the datasets that they train on. In the extreme case, it has been shown that networks can memorize a randomly labeled dataset. In this paper, we propose using the curvature of the loss function around each training sample, averaged over training epochs, as a measure of memorization of a sample. We show that this curvature metric effectively captures memorization statistics, both qualitatively and quantitatively in popular image datasets. We provide quantitative validation of the proposed metric against memorization scores released by Feldman & Zhang (2020). Further, experiments on mislabeled data detection show that corrupted samples are learned with high curvature and using curvature for identifying mislabelled examples outperforms existing approaches. Qualitatively, we find that high curvature samples correspond to long-tailed, mislabeled, or conflicting instances, indicating a likelihood of memorization. Notably, this analysis helps us find, to the best of our knowledge, a novel failure mode on the CIFAR100 and ImageNet datasets: that of duplicated images with differing labels.

Deep Neural Nets (DNNs) have become a pervasive tool for solving many emerging problems. However, they tend to overfit to and memorize the training set. Memorization is of keen interest since it is closely related to several concepts such as generalization, noisy learning, and privacy. To study memorization, Feldman (2019) proposed a formal score, however its computational requirements limit its practical use. Recent research has shown empirical evidence linking input loss curvature (measured by the trace of the loss Hessian w.r.t inputs) and memorization. It was shown to be $\sim3$ orders of magnitude more efficient than calculating the memorization score. However, there is a lack of theoretical understanding linking memorization with input loss curvature. In this paper, we not only investigate this connection but also extend our analysis to establish theoretical links between differential privacy, memorization, and input loss curvature. First, we derive an upper bound on memorization characterized by both differential privacy and input loss curvature. Secondly, we present a novel insight showing that input loss curvature is upper-bounded by the differential privacy parameter. Our theoretical findings are further validated using deep models on CIFAR and ImageNet datasets, showing a strong correlation between our theoretical predictions and results observed in practice.

In this paper, we explore the properties of loss curvature with respect to input data in deep neural networks. Curvature of loss with respect to input (termed input loss curvature) is the trace of the Hessian of the loss with respect to the input. We investigate how input loss curvature varies between train and test sets, and its implications for train-test distinguishability. We develop a theoretical framework that derives an upper bound on the train-test distinguishability based on privacy and the size of the training set. This novel insight fuels the development of a new black box membership inference attack utilizing input loss curvature. We validate our theoretical findings through experiments in computer vision classification tasks, demonstrating that input loss curvature surpasses existing methods in membership inference effectiveness. Our analysis highlights how the performance of membership inference attack (MIA) methods varies with the size of the training set, showing that curvature-based MIA outperforms other methods on sufficiently large datasets. This condition is often met by real datasets, as demonstrated by our results on CIFAR10, CIFAR100, and ImageNet. These findings not only advance our understanding of deep neural network behavior but also improve the ability to test privacy-preserving techniques in machine learning.

The proliferation of big data and analytic workloads has driven the need for cloud compute and cluster-based job processing. With Apache Spark, users can process terabytes of data at ease with hundreds of parallel executors. Providing low latency access to Spark clusters and sessions is a challenging problem due to the large overheads of cluster creation and session startup. In this paper, we introduce Intelligent Pooling, a system for proactively provisioning compute resources to combat the aforementioned overheads. Our system (1) predicts usage patterns using an innovative hybrid Machine Learning (ML) model with low latency and high accuracy; and (2) optimizes the pool size dynamically to meet customer demand while reducing extraneous COGS. The proposed system auto-tunes its hyper-parameters to balance between performance and operational cost with minimal to no engineering input. Evaluated using large-scale production data, Intelligent Pooling achieves up to 43% reduction in cluster idle time compared to static pooling when targeting 99% pool hit rate. Currently deployed in production, Intelligent Pooling is on track to save tens of million dollars in COGS per year as compared to traditional pre-provisioned pools.