`Scale the model, scale the data, scale the compute' is the reigning sentiment in the world of generative AI today. While the impact of model scaling has been extensively studied, we are only beginning to scratch the surface of data scaling and its consequences. This is especially of critical importance in the context of vision-language datasets such as LAION. These datasets are continually growing in size and are built based on large-scale internet dumps such as the Common Crawl, which is known to have numerous drawbacks ranging from quality, legality, and content. The datasets then serve as the backbone for large generative models, contributing to the operationalization and perpetuation of harmful societal and historical biases and stereotypes. In this paper, we investigate the effect of scaling datasets on hateful content through a comparative audit of two datasets: LAION-400M and LAION-2B. Our results show that hate content increased by nearly 12% with dataset scale, measured both qualitatively and quantitatively using a metric that we term as Hate Content Rate (HCR). We also found that filtering dataset contents based on Not Safe For Work (NSFW) values calculated based on images alone does not exclude all the harmful content in alt-text. Instead, we found that trace amounts of hateful, targeted, and aggressive text remain even when carrying out conservative filtering. We end with a reflection and a discussion of the significance of our results for dataset curation and usage in the AI community.Code and the meta-data assets curated in this paper are publicly available at https://github.com/vinayprabhu/hate_scaling. Content warning: This paper contains examples of hateful text that might be disturbing, distressing, and/or offensive.

In this paper, we audit saliency cropping algorithms used by Twitter, Google and Apple to investigate issues pertaining to the male-gaze cropping phenomenon as well as race-gender biases that emerge in post-cropping survival ratios of face-images constituting 3 × 1 grid images. In doing so, we present the first formal empirical study which suggests that the worry of a male-gaze-like image cropping phenomenon on Twitter is not at all far-fetched and it does occur with worryingly high prevalence rates in real-world full-body single-female-subject images shot with logo-littered backdrops. We uncover that while all three saliency cropping frameworks considered in this paper do exhibit acute racial and gender biases, Twitter’s saliency cropping framework uniquely elicits high male-gaze cropping prevalence rates. In order to facilitate reproducing the results presented here, we are open-sourcing both the code and the datasets that we curated at shorturl.at/iuzK9. We hope the computer vision community and saliency cropping researchers will build on the results presented here and extend these investigations to similar frameworks deployed in the real world by other companies such as Microsoft and Facebook.

Accelerometric gait identification systems should ideally be robust to changes in device orientation from the enrollment phase to the deployment phase. However, traditional Convolutional Neural Networks (CNNs) used in these systems compensate poorly for such distributional shifts. In this paper, we target this problem by introducing an SO(3)-equivariant quaternion convolutional kernel inside the CNN. Our architecture (Quaternion CNN) significantly outperforms a traditional CNN in a multi-user gait classification setting. Additionally, the kernels learned by QCNN can be visualized as basis-independent trajectory fragments in Euclidean space, a novel mode of feature visualization and extraction.

In this paper we investigate problematic practices and consequences of large scale vision datasets (LSVDs). We examine broad issues such as the question of consent and justice as well as specific concerns such as the inclusion of verifiably pornographic images in datasets. Taking the ImageNet-ILSVRC-2012 dataset as an example, we perform a cross-sectional model-based quantitative census covering factors such as age, gender, NSFW content scoring, class- wise accuracy, human-cardinality-analysis, and the semanticity of the image class information in order to statistically investigate the extent and subtleties of ethical transgressions. We then use the census to help hand-curate a look-up-table of images in the ImageNet-ILSVRC-2012 dataset that fall into the categories of verifiably pornographic: shot in a non-consensual setting (up-skirt), beach voyeuristic, and exposed private parts. We survey the landscape of harm and threats both the society at large and individuals face due to uncritical and ill-considered dataset curation practices. We then propose possible courses of correction and critique their pros and cons. We have duly open-sourced all of the code and the census meta-datasets generated in this endeavor for the computer vision community to build on. By unveiling the severity of the threats, our hope is to motivate the constitution of mandatory Institutional Review Boards (IRB) for large scale dataset curation.

In this paper, we consider the problem of latent sentiment detection in Online Social Networks such as Twitter. Modeling the underlying social network as an Ising prior, we demonstrate the effect that the underlying social network structure has on the performance of a trivial sentiment detector. In doing so, we introduce a novel communications-oriented framework for characterizing the probability of error and the associated error exponent, based on information theoretic analysis. We study the variation of the calculated error exponent for several stylized network topologies such as the complete network, the star network and the closed-chain network, and show the importance of the network structure in determining detection performance.

This paper considers the problem of filter design with secrecy constraints, where two legitimate parties (Alice and Bob) communicate in the presence of an eavesdropper (Eve), over a Gaussian multiple-input multiple-output (MIMO) wiretap channel. This problem involves the design of transmit and receive filters which minimize the mean-square error (MSE) between the legitimate parties, whilst assuring that the eavesdropper MSE remains above a certain level. We characterize the form of the optimal transmit filter when both the legitimate receiver and the eavesdropper employ Zero-Forcing (ZF) filters. By capitalizing on the dual problem, we also show that the original matrix optimization problem can be reduced to a simple scalar optimization problem, whose solution can be readily computed by employing a simple bisection method. Numerical results illustrate the main conclusions.

In this correspondence, we characterize the probability of secrecy-outage and the asymptotic high-signal-to-noise ratio (SNR) ε-outage secrecy-capacity for a single-input-single-output-multi-eavesdropper (SISOME) wireless system with eavesdroppers performing maximum ratio combining (MRC) or selection diversity combining (SDC) reception. We also consider the SISO2E case with eavesdropper antenna-correlation and finally, analyze the scenario where the eavesdropper has Rician fading links with the transmitter.

In this paper, we consider secure communications between a single antenna transmitter and a single antenna receiver in the presence of a multiple antenna eavesdropper employing MRC or SDC reception. We show that an M-antenna eavesdropper performing SDC reception has precisely the same degrading effect on secrecy as M single antenna eavesdroppers under independent fading conditions. We also show that an M-antenna eavesdropper performing MRC reception causes greater secrecy degradation than an M-antenna eavesdropper performing SDC reception or M single antenna eavesdroppers. We also derive closed-form expressions for the asymptotic high SNR outage secrecy capacity for both the MRC and the SDC eavesdropper cases. Various numerical results are presented that corroborate the analysis.

In the paper, we propose a conditionally biased linear estimator to estimate a deterministic vector h subjected to a linear transformation M and corrupted by zero-mean additive white Gaussian noise, w. We begin by drawing inspiration from the rank shaping idea of the Rank Shaped Least Squares (RSLS) estimator. We then incorporate the estimator into the Covariance Shaping Least Squares (CSLS) framework, which results in inheritance of attractive properties, such as achieving the Cramer-Rao Lower Bound for biased estimators (B-CRLB) and conditional dominance over the Least Squares Estimator (LSE). We further introduce a bias control parameter that not only allows elegant control of the trade-off between the bias and the MSE floor, but also helps in elucidating the common framework that binds the proposed estimator, RSLS, CSLS and LS estimators. We conclude by applying the proposed estimator for pilot-aided channel estimation in OFDM systems.

The exact time-of-arrival (TOA) information of the multi-path signals is crucial for optimal channel estimation in vehicular OFDM systems. Super-resolution algorithms such as ESPRIT (Estimation of Signal Parameters via Rotational Invariance Technique) have been applied to retrieve this information from embedded pilots in the received OFDM symbols. These algorithms have high applicability for vehicular wireless environs characterized by high Doppler frequencies which leads to faster auto-correlation averaging. In this paper, we propose to improve upon the classical ESPRIT algorithm by incorporating the iterative reduced rank Hankel approximation (RRHA) technique. We show that this allows achieving of a lower MSE (mean square error) compared to the classical ESPRIT algorithm. Further, we portray that the degree of improvement brought forth increases with the increased number of RRHA-iterations.