Siddharth Krishna Kumar

Siddharth Krishna Kumar

This paper critically examines the fundamental distinctions between gradient methods applied to non-differentiable functions (NGDMs) and classical gradient descents (GDs) for differentiable functions, revealing significant gaps in current deep learning optimization theory. We demonstrate that NGDMs exhibit markedly different convergence properties compared to GDs, strongly challenging the applicability of extensive neural network convergence literature based on $L-smoothness$ to non-smooth neural networks. Our analysis reveals paradoxical behavior of NDGM solutions for $L_{1}$-regularized problems, where increasing regularization counterintuitively leads to larger $L_{1}$ norms of optimal solutions. This finding calls into question widely adopted $L_{1}$ penalization techniques for network pruning. We further challenge the common assumption that optimization algorithms like RMSProp behave similarly in differentiable and non-differentiable contexts. Expanding on the Edge of Stability phenomenon, we demonstrate its occurrence in a broader class of functions, including Lipschitz continuous convex differentiable functions. This finding raises important questions about its relevance and interpretation in non-convex, non-differentiable neural networks, particularly those using ReLU activations. Our work identifies critical misunderstandings of NDGMs in influential literature, stemming from an overreliance on strong smoothness assumptions. These findings necessitate a reevaluation of optimization dynamics in deep learning, emphasizing the crucial need for more nuanced theoretical foundations in analyzing these complex systems.

Rutvik Vijjali, Prathyush Potluri, Siddharth Kumar et al.

The rapid advancement of technology in online communication via social media platforms has led to a prolific rise in the spread of misinformation and fake news. Fake news is especially rampant in the current COVID-19 pandemic, leading to people believing in false and potentially harmful claims and stories. Detecting fake news quickly can alleviate the spread of panic, chaos and potential health hazards. We developed a two stage automated pipeline for COVID-19 fake news detection using state of the art machine learning models for natural language processing. The first model leverages a novel fact checking algorithm that retrieves the most relevant facts concerning user claims about particular COVID-19 claims. The second model verifies the level of truth in the claim by computing the textual entailment between the claim and the true facts retrieved from a manually curated COVID-19 dataset. The dataset is based on a publicly available knowledge source consisting of more than 5000 COVID-19 false claims and verified explanations, a subset of which was internally annotated and cross-validated to train and evaluate our models. We evaluate a series of models based on classical text-based features to more contextual Transformer based models and observe that a model pipeline based on BERT and ALBERT for the two stages respectively yields the best results.

Siddharth Krishna Kumar

It is well known that a determined adversary can fool a neural network by making imperceptible adversarial perturbations to an image. Recent studies have shown that these perturbations can be detected even without information about the neural network if the strategy taken by the adversary is known beforehand. Unfortunately, these studies suffer from the generalization limitation -- the detection method has to be recalibrated every time the adversary changes his strategy. In this study, we attempt to overcome the generalization limitation by deriving a metric which reliably identifies adversarial images even when the approach taken by the adversary is unknown. Our metric leverages key differences between the spectra of clean and adversarial images when an image is treated as a matrix. Our metric is able to detect adversarial images across different datasets and attack strategies without any additional re-calibration. In addition, our approach provides geometric insights into several unanswered questions about adversarial perturbations.

Siddharth Krishna Kumar

A proper initialization of the weights in a neural network is critical to its convergence. Current insights into weight initialization come primarily from linear activation functions. In this paper, I develop a theory for weight initializations with non-linear activations. First, I derive a general weight initialization strategy for any neural network using activation functions differentiable at 0. Next, I derive the weight initialization strategy for the Rectified Linear Unit (RELU), and provide theoretical insights into why the Xavier initialization is a poor choice with RELU activations. My analysis provides a clear demonstration of the role of non-linearities in determining the proper weight initializations.