Abstract: Human gaze behavior is a fundamental cue for understanding social intent, human-machine interaction, and cognitive processes. This thesis addresses the challenges of gaze target estimation (GTE), also known as gaze following, by developing a holistic understanding of gaze in complex environments.

The first part of this work improves GTE performance by introducing Patch-level Distribution Prediction (PDP). Unlike traditional models that rely on strict pixel-wise regression, PDP models gaze as a distribution over patches, which better accounts for annotation variance and bridges the gap between target location and in/out-of-frame prediction. To address the laborious nature of data labeling, the second part presents GCDR, the first semi-supervised method for gaze following. By prompting large Visual Question Answering (VQA) models to generate initial Grad-CAM heatmaps and refining them with a diffusion model, this method achieves high performance with significantly fewer human annotations. The third part expands the applicability of GTE to multi-camera environments. By introducing the Multi-View Gaze Target (MVGT) dataset, along with two novel frameworks for integrating information between multiple views and predicting the gaze target across views, we explore a new direction that overcomes single-view limitations such as face occlusion and out-of-view targets.

Building on these foundations, the final part of this thesis proposes a new direction toward semantic social gaze understanding using next-generation multimodal Large Language Models (LLMs). Rather than focusing solely on geometric gaze target localization, we aim to enrich gaze prediction with semantic and relational interpretation in complex social scenes. To this end, we will leverage existing gaze following datasets to derive social gaze supervision, including mutual gaze and shared attention, and obtain aligned language descriptions of scene-level gaze behaviors. This proposed work will enable the model to not only locate gaze targets but also predict structured social gaze relations among individuals, meanwhile generating a concise natural-language summary describing the dominant gaze interactions. By integrating spatial gaze estimation, social relation reasoning, and language-based scene understanding within a unified multimodal model, this work takes an important step toward a holistic understanding of human gaze behavior in real-world environments.

Speaker: Qiaomu Miao
TITLE: Towards a Theory of Encode/Decoder Architectures by Andrej Risteski of CMU

ABSTRACT: A common choice of architecture in representation learning (i.e., learning a good embedding of the data) is an encoder/decoder architecture, which tries to map a part of the input into a good latent representation (via an encoder), and predict the remaining part of the input (via a decoder). Two common examples are universal machine translation: where one tries to learn to translate between any pair of a set of languages via a common latent language, given paired up corpora for only a part of the pairs; and contextual encoders -- where one tries to predict a part of the image, given the rest of the image.
 
We will give a framework for analyzing the sample complexity of such architectures -- i.e., how many pairs of languages do we need to have paired up corpora for? How many image prediction tasks do we have to solve to get a good representation?

You are cordially invited to attend the biweekly Brookhaven AI Mixer (BAM). BAM includes one short talk on AI research happening at BNL, followed by an open mixer over coffee and snacks for everyone to network and discuss all things AI. The first half hour will consist of presentations that will be available via ZOOM, and the second half hour will be for in person only networking.

Join us every other Tuesday at noon in CDSD's Training Room (building 725, 2nd floor) to learn about interesting AI methods and applications, engage with potential collaborators, prepare for pending FASST funding calls, and build a community of AI for Science at BNL.

HPCortex - a new, general-purpose machine learning library for HPC

Abstract: I will introduce HPCortex, a lightweight, C++, MPI-native machine-learning library for heterogeneous HPC systems. It implements many common architecture patterns including transformers, graph neural networks, and convolutional networks, and delivers performance portability across NVIDIA, AMD, and Intel GPUs while depending only on MPI and standard compiler/BLAS stacks. I will illustrate its capabilities via a surrogate model for the RHIC AGS Booster digital twin, a simple GNN for a coupled spring system, and a compact language model, then outline the roadmap.

Biography: Christopher is a research scientist and head of the Scientific Computing Applications Group in the Computational Science Department at Brookhaven National Laboratory. Previously he was an assistant staff scientist in the Physics Dept. at Columbia University, and held physics postdoctoral research positions at both Brookhaven and Columbia. He earned his Ph.D in Theoretical Physics from the University of Edinburgh, UK.
His scientific background is in lattice QCD and high performance computing, but since joining Brookhaven in 2020 his research interests have expanded to include machine learning, applied mathematics and performance analysis, with a particular emphasis on building tools to support scientific research on HPC systems.

Location: CDS, Bldg. 725, Training Room

Join ZoomGov Meeting: https://bnl.zoomgov.com/j/1604143373?pwd=hHT2yaIjahBIQ6tieURFqs8Pwex9gU.1

Meeting ID: 160 414 3373
Passcode: 277410

Face Editing with Machine Learning presented by Zhixin Shu

ABSTRACT: The face is the most informative feature of humans and has been a long-standing research topic in Computer Vision and Graphics. Images of faces are also ubiquitous in photography and social media, and people have devoted significant resources to capturing and editing face images. Face editing can be broadly viewed as the encoding, manipulation and the decoding of some representations for face images. The challenges are that we want to manipulate an image in a controllable way and generate results that are both desirable and as realistic as possible. This thesis explores different Machine Learning-based face-editing approaches. I discuss the role of machine learning for achieving desirable edits by learning both the physical aspects as well as the statistical manifold of human faces. In my work for eye-editing, I discuss the importance of understanding multiple physical elements of a face image, such as shape, illumination, pose, etc. In a deep-learning-based approach, I introduce image formation domain knowledge to the construction and training of a neural network. This network provides transparent access to the disentangled representations of the aforementioned physical properties. With this network, we can achieve various face editing tasks in forms of representation manipulation. After that, I introduce Deforming Autoencoders, a network that learns to disentangle shape and appearance in an unsupervised manner. This disentanglement is beneficial for the learning of some other factors of variations, such as illumination and facial expression. In an extension of Deforming Autoencoders, we incorporate non-rigid structure-from-motion to learn a 3D morphable model for faces that only requires an image set for training. At last, I describe an image-to-image network for 3D face reconstruction, which also utilizes structure-from-motion in deep learning. With real face images in training, this network not only reconstructs 3D faces more accurately than prior art but also has better generalization ability in real-life testing cases.

You are cordially invited to attend the biweekly Brookhaven AI Mixer (BAM). BAM includes one short talk on AI research happening at BNL, followed by an open mixer. The first half hour will consist of presentations that will be available via ZOOM, and the second half hour will be for in person only networking.

We meet every other Tuesday at noon in CDSD's Training Room (building 725, room 2-124) to learn about interesting AI methods and applications, engage with potential collaborators, prepare for pending FASST funding calls, and build a community of AI for Science at BNL.

Abstract: Identifying model Hamiltonians is a vital step toward creating predictive models of materials. W​e combined Bayesian optimization with the EDRIXS numerical package to infer Hamiltonian parameters from resonant inelastic X-ray scattering (RIXS) spectra within the single atom approximation. To evaluate the efficacy of our method, we tested it on experimental RIXS spectra for several materials and demonstrated that it can reproduce results obtained from hand-fitted parameters to a precision similar to expert human analysis while providing a more systematic mapping of parameter space. Our work provides a key first step toward solving the inverse scattering problem to extract effective multi- orbital models from information-dense RIXS measurements, which can be applied to a host of quantum materials.

Biography: Marton Lajer is a postdoctoral researcher at the Condensed Matter Physics and Materials Science Department, Brookhaven National Laboratory. Marton​ obtained his PhD in theoretical physics at the Eotvos Lorand University, Hungary, in 2021. He was a​ junior research fellow at the Wigner Research Centre for Physics in Budapest before joining BNL in September 2022. His background spans various analytical and performance-critical numerical methods, mostly in the context of low- dimensional quantum field theories and quantum many-body systems. His research currently focuses on incorporating AI-enhanced methods to various problems in inelastic spectroscopy.

In addition to our speaker, we will have a number of CDS staff in attendance with expertise in AI methods and applications including image analysis, foundation models development, and inverse problem solving.

Location: CDS, Bldg. 725, Training Room

Join ZoomGov Meeting: https://bnl.zoomgov.com/j/1604383624?pwd=ffQ5cUPNxTI7nzClKQO6cnsNbhF9Vf.1

Meeting ID: 160 438 3624
Passcode: 558449

Please Note: Due to a funding shortfall, we are for the time being no longer able to provide pizza and sodas for these events. We will have coffee though, and all are of course welcome to bring their lunch.



Place:  https://stonybrook.zoom.us/j/99167126152?pwd=TFpEYzM0aFhiOFJxSFJEb1JSS3YyQT09  

Time: 3 PM EST - Dec, 16th, 2020 

Abstract: 

Shadows provide useful cues to analyze visual scenes but also hamper many computer vision algorithms such as image segmentation, object detection, or tracking. For those reasons, shadow detection and shadow removal have been well-studied in computer vision.

Early work on shadow detection and removal focused on physical illumination models of shadows. These methods can express, identify, and remove shadows in a physically plausible manner. However, these models are often hard to optimize and are slow during inference due to their reliance on hand-designed image features. Recently, deep-learning approaches have achieved breakthroughs in performance for both shadow detection and removal. They learn to extract useful features through training while being extremely efficient during inference. However, these models are data-dependent, opaque, and ignore the physical aspects of shadows. Thus they often lack generalization and produce inconsistent results.

We propose incorporating physical illumination constraints of shadows into deep-learning models. These constraints force the networks to more closely follow the physics of shadows, enabling them to systematically and realistically modify shadows in images. For shadow detection, we present a novel Generative Adversarial Network (GAN) based model where the generator learns to generate images with realistic attenuated shadows that can be used to train a shadow detector. For shadow removal, we propose a method that uses deep-networks to estimate the unknown parameters of a shadow image formation model that removes shadows. The system outputs high-quality shadow-free images with little or no image artifacts and achieves state-of-the-art performance in shadow removal when trained on a fully-supervised setting. Moreover, the system is easy to train and constrain since the shadow removal mapping is strictly defined by the simplified illumination model with interpretable parameters. Thus, it can be trained even with a much weaker form of supervision signal. In particular, we show that we can use two sets of patches, shadow and shadow-free, to train our shadow decomposition framework via an adversarial system. These patches are cropped from the shadow images themselves.
Therefore, this is the first deep-learning method for shadow removal that can be trained without any shadow-free images, providing an alternative solution to the paired data dependency issue. The advantage of this training scheme is even more pronounced when tested on a novel domain such as video shadow removal where the method can be fine-tuned on a testing video with only the shadow masks generated by a pre-trained shadow detector and further improves shadow removal results.