Data in the real world is commonly imbalanced across classes. Training neural networks on imbalanced datasets often leads to poor performance on rare classes. Existing work in this area has primarily focused on Convolution Neural Networks (CNN), which are increasingly being replaced by Self-Attention-based Vision Transformers (ViT). Fundamentally, ViTs differ from CNNs in that they offer the flexibility in learning the appropriate inductive bias conducive to improving performance. This work is among the first to evaluate the performance of ViTs under class imbalance. We find that accuracy degradation in the presence of class imbalance is much more prominent in ViTs compared to CNNs. This degradation can be partially mitigated through loss reweighting - a popular strategy that increases the loss contributed by rare classes. We investigate the impact of loss reweighting on different components of a ViT, namely, the patch embedding, self-attention backbone, and linear classifier. Our ongoing investigations reveal that loss reweighting impacts mostly the linear classifier and self-attention backbone while having a small and negligible effect on the embedding layer.

We present ConceptEvo, a unified interpretation framework for deep neural networks (DNNs) that reveals the inception and evolution of learned concepts during training. Our work addresses a critical gap in DNN interpretation research, as existing methods primarily focus on post-training interpretation. ConceptEvo introduces two novel technical contributions: (1) an algorithm that generates a unified semantic space, enabling side-by-side comparison of different models during training, and (2) an algorithm that discovers and quantifies important concept evolutions for class predictions. Through a large-scale human evaluation and quantitative experiments, we demonstrate that ConceptEvo successfully identifies concept evolutions across different models, which are not only comprehensible to humans but also crucial for class predictions. ConceptEvo is applicable to both modern DNN architectures, such as ConvNeXt, and classic DNNs, such as VGGs and InceptionV3.

Computer vision is playing an increasingly important role in automated malware detection with the rise of the image-based binary representation. These binary images are fast to generate, require no feature engineering, and are resilient to popular obfuscation methods. Significant research has been conducted in this area, however, it has been restricted to small-scale or private datasets that only a few industry labs and research teams have access to. This lack of availability hinders examination of existing work, development of new research, and dissemination of ideas. We release MalNet-Image, the largest public cybersecurity image database, offering 24x more images and 70x more classes than existing databases (available at https://mal-net.org). MalNet-Image contains over 1.2 million malware images-across 47 types and 696 families---democratizing image-based malware capabilities by enabling researchers and practitioners to evaluate techniques that were previously reported in propriety settings. We report the first million-scale malware detection results on binary images. MalNet-Image unlocks new and unique opportunities to advance the frontiers of machine learning, enabling new research directions into vision-based cyber defenses, multi-class imbalanced classification, and interpretable security.

Existing research on making sense of deep neural networks often focuses on neuron-level interpretation, which may not adequately capture the bigger picture of how concepts are collectively encoded by multiple neurons. We present Neurocartography, an interactive system that scalably summarizes and visualizes concepts learned by neural networks. It automatically discovers and groups neurons that detect the same concepts, and describes how such neuron groups interact to form higher-level concepts and the subsequent predictions. Neurocartography introduces two scalable summarization techniques: (1) neuron clustering groups neurons based on the semantic similarity of the concepts detected by neurons (e.g., neurons detecting "dog faces" of different breeds are grouped); and (2) neuron embedding encodes the associations between related concepts based on how often they co-occur (e.g., neurons detecting "dog face" and "dog tail" are placed closer in the embedding space). Key to our scalable techniques is the ability to efficiently compute all neuron pairs' relationships, in time linear to the number of neurons instead of quadratic time. Neurocartography scales to large data, such as the ImageNet dataset with 1.2M images. The system's tightly coordinated views integrate the scalable techniques to visualize the concepts and their relationships, projecting the concept associations to a 2D space in Neuron Projection View, and summarizing neuron clusters and their relationships in Graph View. Through a large-scale human evaluation, we demonstrate that our technique discovers neuron groups that represent coherent, human-meaningful concepts. And through usage scenarios, we describe how our approaches enable interesting and surprising discoveries, such as concept cascades of related and isolated concepts. The Neurocartography visualization runs in modern browsers and is open-sourced.

Class imbalance is a significant i ssue t hat causes neural networks to underfit t o t he r are c lasses. Traditional mitigation strategies include loss reshaping and data resampling which amount to increasing the loss contribution of minority classes and decreasing the loss contributed by the majority ones. However, by treating each example within a class equally, these methods lead to undesirable scenarios where hard-to-classify examples from the majority classes are down-weighted and easy-to-classify examples from the minority classes are up-weighted. We propose the Hardness Aware Reweighting (HAR) framework, which circumvents this issue by increasing the loss contribution of hard examples from both the majority and minority classes. This is achieved by augmenting a neural network with intermediate classifier b ranches t o e nable e arly-exiting d uring t raining. Experimental results on large-scale datasets demonstrate that HAR consistently improves state-of-the-art accuracy while saving up to 20% of inference FLOPS.

We propose CUP, a new method for compressing and accelerating deep neural networks. At its core, CUP achieves compression by clustering and pruning similar filters in each layer. For clustering, CUP uses hierarchical clustering which allows for an elegant parameterization of model capacity through a single hyper-parameter t. We observe that by increasing t, CUP can dynamically reduce model capacity through non-uniform layer-wise pruning leading to two advantages. First, CUP can effectively compress a model to within the desired compute budget through a simple line-search on t. Second, through a simple extension, CUP can obtain the pruned model in a single training pass leading to large savings in training time. On Imagenet, CUP leads to a 2.47× FLOPS reduction on Resnet-50 with less than 1% drop in top-5 accuracy. Notably, in the retrain-free setting, CUP-RF saves over 10 hours of training time on 3 GPUs, in comparison to state-of-the-art methods. The code for CUP is open sourced1.

We tackle the problem of visual search under resource constraints. Existing systems use the same embedding model to compute representations (embeddings) for the query and gallery images. Such systems inherently face a hard accuracy-efficiency trade-off: the embedding model needs to be large enough to ensure high accuracy, yet small enough to enable query-embedding computation on resource-constrained platforms. This trade-off could be mitigated if gallery embeddings are generated from a large model and query embeddings are extracted using a compact model. The key to building such a system is to ensure representation compatibility between the query and gallery models. In this paper, we address two forms of compatibility: One enforced by modifying the parameters of each model that computes the embeddings. The other by modifying the architectures that compute the embeddings, leading to compatibility-aware neural architecture search (Cmp-NAS). We test Cmp-NAS on challenging retrieval tasks for fashion images (DeepFashion2), and face images (IJB-C). Compared to ordinary (homogeneous) visual search using the largest embedding model (paragon), Cmp-NAS achieves 80-fold and 23-fold cost reduction while maintaining accuracy within 0.3% and 1.6% of the paragon on DeepFashion2 and IJB-C respectively.

In recent years, significant attention has been devoted towards integrating deep learning technologies in the healthcare domain. However, to safely and practically deploy deep learning models for home health monitoring, two significant challenges must be addressed: the models should be (1) robust against noise; and (2) compact and energy-efficient. We propose Rest , a new method that simultaneously tackles both issues via 1) adversarial training and controlling the Lipschitz constant of the neural network through spectral regularization while 2) enabling neural network compression through sparsity regularization. We demonstrate that Rest produces highly-robust and efficient models that substantially outperform the original full-sized models in the presence of noise. For the sleep staging task over single-channel electroencephalogram (EEG), the Rest model achieves a macro-F1 score of 0.67 vs. 0.39 achieved by a state-of-the-art model in the presence of Gaussian noise while obtaining 19 × parameter reduction and 15 × MFLOPS reduction on two large, real-world EEG datasets. By deploying these models to an Android application on a smartphone, we quantitatively observe that Rest allows models to achieve up to 17 × energy reduction and 9 × faster inference. We open source the code repository with this paper: https://github.com/duggalrahul/REST.