Search

In set-based face recognition, we aim to compute the most discriminative descriptor from an unbounded set of images and videos showing a single person. A discriminative descriptor balances two policies when aggregating information from a given set. The first is a quality-based policy: emphasizing high-quality and down-weighting low-quality images. The second is a diversity-based policy: emphasizing unique images in the set and down-weighting multiple occurrences of similar images as found in video clips which can overwhelm the set representation. This work frames face-set representation as a differentiable coreset selection problem. Our model learns how to select a small coreset of the input set that balances quality and diversity policies using a learned metric parameterized by the face quality, optimized end-to-end. The selection process is a differentiable farthest-point sampling (FPS) realized by approximating the non-differentiable Argmax operation with differentiable sampling from the Gumbel-Softmax distribution of distances. The small coreset is later used as queries in a self and cross-attention architecture to enrich the descriptor with information from the whole set. Our model is order-invariant and linear in the input set size. We set a new SOTA to set face verification on the IJB-B and IJB-C datasets. Our code is publicly available., Comment: Accepted to AAAI-24

Facial landmarks (FLM) estimation is a critical component in many face-related applications. In this work, we aim to optimize for both accuracy and speed and explore the trade-off between them. Our key observation is that not all faces are created equal. Frontal faces with neutral expressions converge faster than faces with extreme poses or expressions. To differentiate among samples, we train our model to predict the regression error after each iteration. If the current iteration is accurate enough, we stop iterating, saving redundant iterations while keeping the accuracy in check. We also observe that as neighboring patches overlap, we can infer all facial landmarks (FLMs) with only a small number of patches without a major accuracy sacrifice. Architecturally, we offer a multi-scale, patch-based, lightweight feature extractor with a fine-grained local patch attention module, which computes a patch weighting according to the information in the patch itself and enhances the expressive power of the patch features. We analyze the patch attention data to infer where the model is attending when regressing facial landmarks and compare it to face attention in humans. Our model runs in real-time on a mobile device GPU, with 95 Mega Multiply-Add (MMA) operations, outperforming all state-of-the-art methods under 1000 MMA, with a normalized mean error of 8.16 on the 300W challenging dataset., Comment: Accepted to the 29th ACM International Conference on Multimedia (MM 21)

The 2D Least Median of Squares (LMS) is a popular tool in robust regression because of its high breakdown point: up to half of the input data can be contaminated with outliers without affecting the accuracy of the LMS estimator. The complexity of 2D LMS estimation has been shown to be $\Omega(n^2)$ where $n$ is the total number of points. This high theoretical complexity along with the availability of graphics processing units (GPU) motivates the development of a fast, parallel, GPU-based algorithm for LMS computation. We present a CUDA based algorithm for LMS computation and show it to be much faster than the optimal state of the art single threaded CPU algorithm. We begin by describing the proposed method and analyzing its performance. We then demonstrate how it can be used to modify the well-known Hough Transform (HT) in order to efficiently detect image lines in noisy images. Our method is compared with standard HT-based line detection methods and shown to overcome their shortcomings in terms of both efficiency and accuracy.

In this project, I study the effect of subjective beliefs about HIV infection on fertility decisions in a context of high HIV prevalence and evaluate the impact of different policy interventions, such as HIV testing programs, informational campaigns, and antiretroviral therapy distribution, on fertility and child mortality. I develop a dynamic discrete-choice life-cycle fertility model in which expectations about the life horizon and child survival depend on a perceived HIV infection hazard process, which is allowed to differ from the actual hazard. In the model, women form and update beliefs about their HIV status and about their own and their children's survival in future periods. Women also update their beliefs when their HIV status is revealed by an HIV test. Model parameters are estimated by maximum likelihood with longitudinal data from the Malawi Diffusion and Ideational Change Project, which contain family rosters, information on HIV testing, and measures of subjective beliefs about own HIV status. I use the model to evaluate the effect of HIV on fertility by simulating behavior in an environment without HIV. The model successfully fits the fertility patterns in the data, as well as the distribution of reported beliefs about own HIV status. Results show that the presence of HIV reduces the average number of births a woman has during her life-cycle by 0.23. Although this effect is due both to shorter life-span of women infected with HIV as well as to a behavioral effect, the behavioral effect is the dominant mechanism that reduces fertility. I find that HIV testing can be effective at reducing fertility of infected women, leading to a reduction of child mortality and orphanhood. I also find that although provision of antiretroviral treatment that prevents mother to child transmission will reduce child mortality, it has negligible effect on the average number of life-cycle births.

Thesis (S.M.)--Massachusetts Institute of Technology, Sloan School of Management, 2004., Includes bibliographical references., The primary goal of this research effort was to integrate several pieces of relatively simple dynamics simulations into a full blown system dynamics simulation of the telecom industry. The economic simulation model was used to analyze potential dynamics, reaching basic qualitative insights on the forces behind the industry. Examples of the pieces which comprised the model include the product adoption life cycle, product substitution models, stochastic R&D investment models, as well as an economic model for determining market share under price competition. The work focused on key attributes and areas of the telecom industry, including network externalities, investment strategies, regulation, standardization of products, and behavior of consumer welfare. Main conclusions include the convexity of risk/return tradeoff in R&D investment under severe network externalities, as well as substitution as a matching device for innovation clockspeed in competing industries. Accordingly, it is shown that mismatch in innovation clockspeed, pricing clockspeed or product introduction timing may lead to an unsustainable competitive position. The work shows the advantages of R&D competition in driving consumer welfare. A theoretical framework has been put in place for considering the dynamic influence of regulation and standardization on investment incentives and choices. Through this framework, we can consider regulation and standardization effects in short-term and long-term, and how these correspond to the competitive stability of the industry., by Gil Shapira., S.M.