Search

We study the momentum-based minimization of a diffuse perimeter functional on Euclidean spaces and on graphs with applications to semi-supervised classification tasks in machine learning. While the gradient flow in the task at hand is a parabolic partial differential equation, the momentum-method corresponds to a damped hyperbolic PDE, leading to qualitatively and quantitatively different trajectories. Using a convex-concave splitting-based FISTA-type time discretization, we demonstrate empirically that momentum can lead to faster convergence if the time step size is large but not too large. With large time steps, the PDE analysis offers only limited insight into the geometric behavior of solutions and typical hyperbolic phenomena like loss of regularity are not be observed in sample simulations.

Due to their large sizes, volumetric scans and whole-slide pathology images (WSIs) are often processed by extracting embeddings from local regions and then an aggregator makes predictions from this set. However, current methods require post-hoc visualization techniques (e.g., Grad-CAM) and often fail to localize small yet clinically crucial details. To address these limitations, we introduce INSIGHT, a novel weakly-supervised aggregator that integrates heatmap generation as an inductive bias. Starting from pre-trained feature maps, INSIGHT employs a detection module with small convolutional kernels to capture fine details and a context module with a broader receptive field to suppress local false positives. The resulting internal heatmap highlights diagnostically relevant regions. On CT and WSI benchmarks, INSIGHT achieves state-of-the-art classification results and high weakly-labeled semantic segmentation performance. Project website and code are available at: https://zhangdylan83.github.io/ewsmia/

This study introduces SLLMBO, an innovative framework leveraging large language models (LLMs) for hyperparameter optimization (HPO), incorporating dynamic search space adaptability, enhanced parameter space exploitation, and a novel LLM-tree-structured parzen estimator (LLM-TPE) sampler. By addressing limitations in recent fully LLM-based methods and traditional bayesian optimization (BO), SLLMBO achieves more robust optimization. This comprehensive benchmarking evaluates multiple LLMs, including GPT-3.5-Turbo, GPT-4o, Claude-Sonnet-3.5, and Gemini-1.5-Flash, extending prior work and establishing SLLMBO as the first framework to benchmark a diverse set of LLMs for HPO. By integrating LLMs' established strengths in parameter initialization with the exploitation abilities demonstrated in this study, alongside TPE's exploration capabilities, the LLM-TPE sampler achieves a balanced exploration-exploitation trade-off, reduces API costs, and mitigates premature early stoppings for more effective parameter searches. Across 14 tabular tasks in classification and regression, the LLM-TPE sampler outperformed fully LLM-based methods and achieved superior results over BO methods in 9 tasks. Testing early stopping in budget-constrained scenarios demonstrated competitive performance, indicating that LLM-based methods generally benefit from extended iterations for optimal results. This work lays the foundation for future research exploring open-source LLMs, reproducibility of LLM results in HPO, and benchmarking SLLMBO on complex datasets, such as image classification, segmentation, and machine translation.

Generative large language models (LLMs) exhibit impressive capabilities, which can be further augmented by integrating a pre-trained vision model into the original LLM to create a multimodal LLM (MLLM). However, this integration often significantly decreases performance on natural language understanding and generation tasks, compared to the original LLM. This study investigates this issue using the LLaVA MLLM, treating the integration as a continual learning problem. We evaluate five continual learning methods to mitigate forgetting and identify a technique that enhances visual understanding while minimizing linguistic performance loss. Our approach reduces linguistic performance degradation by up to 15\% over the LLaVA recipe, while maintaining high multimodal accuracy. We also demonstrate the robustness of our method through continual learning on a sequence of vision-language tasks, effectively preserving linguistic skills while acquiring new multimodal capabilities., Comment: NeurIPS 2024 Workshop on Scalable Continual Learning for Lifelong Foundation Models

While momentum-based optimization algorithms are commonly used in the notoriously non-convex optimization problems of deep learning, their analysis has historically been restricted to the convex and strongly convex setting. In this article, we partially close this gap between theory and practice and demonstrate that virtually identical guarantees can be obtained in optimization problems with a `benign' non-convexity. We show that these weaker geometric assumptions are well justified in overparametrized deep learning, at least locally. Variations of this result are obtained for a continuous time model of Nesterov's accelerated gradient descent algorithm (NAG), the classical discrete time version of NAG, and versions of NAG with stochastic gradient estimates with purely additive noise and with noise that exhibits both additive and multiplicative scaling.

The domain of laser fusion presents a unique and challenging predictive modeling application landscape for machine learning methods due to high problem complexity and limited training data. Data-driven approaches utilizing prescribed functional forms, inductive biases and physics-informed learning (PIL) schemes have been successful in the past for achieving desired generalization ability and model interpretation that aligns with physics expectations. In complex multi-physics application domains, however, it is not always obvious how architectural biases or discriminative penalties can be formulated. In this work, focusing on nuclear fusion energy using high powered lasers, we present the use of Kolmogorov-Arnold Networks (KANs) as an alternative to PIL for developing a new type of data-driven predictive model which is able to achieve high prediction accuracy and physics interpretability. A KAN based model, a MLP with PIL, and a baseline MLP model are compared in generalization ability and interpretation with a domain expert-derived symbolic regression model. Through empirical studies in this high physics complexity domain, we show that KANs can potentially provide benefits when developing predictive models for data-starved physics applications.

Rapid determination of molecular structures can greatly accelerate workflows across many chemical disciplines. However, elucidating structure using only one-dimensional (1D) NMR spectra, the most readily accessible data, remains an extremely challenging problem because of the combinatorial explosion of the number of possible molecules as the number of constituent atoms is increased. Here, we introduce a multitask machine learning framework that predicts the molecular structure (formula and connectivity) of an unknown compound solely based on its 1D 1H and/or 13C NMR spectra. First, we show how a transformer architecture can be constructed to efficiently solve the task, traditionally performed by chemists, of assembling large numbers of molecular fragments into molecular structures. Integrating this capability with a convolutional neural network (CNN), we build an end-to-end model for predicting structure from spectra that is fast and accurate. We demonstrate the effectiveness of this framework on molecules with up to 19 heavy (non-hydrogen) atoms, a size for which there are trillions of possible structures. Without relying on any prior chemical knowledge such as the molecular formula, we show that our approach predicts the exact molecule 69.6% of the time within the first 15 predictions, reducing the search space by up to 11 orders of magnitude.

With the advent of multi-modal large language models (MLLMs), datasets used for visual question answering (VQA) and referring expression comprehension have seen a resurgence. However, the most popular datasets used to evaluate MLLMs are some of the earliest ones created, and they have many known problems, including extreme bias, spurious correlations, and an inability to permit fine-grained analysis. In this paper, we pioneer evaluating recent MLLMs (LLaVA 1.5, LLaVA-NeXT, BLIP2, InstructBLIP, GPT-4V, and GPT-4o) on datasets designed to address weaknesses in earlier ones. We assess three VQA datasets: 1) TDIUC, which permits fine-grained analysis on 12 question types; 2) TallyQA, which has simple and complex counting questions; and 3) DVQA, which requires optical character recognition for chart understanding. We also study VQDv1, a dataset that requires identifying all image regions that satisfy a given query. Our experiments reveal the weaknesses of many MLLMs that have not previously been reported. Our code is integrated into the widely used LAVIS framework for MLLM evaluation, enabling the rapid assessment of future MLLMs. Project webpage: https://kevinlujian.github.io/MLLM_Evaluations/

Measuring the properties of the intergalactic medium (IGM) and sources during the Epoch of Reionization (EoR) is of immense importance. We explore the prospects of probing the IGM and sources through redshifted 21-cm observations of individual ionized bubbles surrounding known luminous sources during the EoR. Accordingly, we simulate HI 21-cm maps, foreground contaminants, and system noise which are specific to the uGMRT and SKA1-Low observations. Following the subtraction of the foreground from the total visibility, we employ a visibility-based matched filter technique to optimally combine the desired HI 21-cm signal while minimizing the system noise. Our analysis suggests that these ionized bubbles can be detected with more than $5 \sigma$ significance using approximately $\sim 2000$ and $\sim 3000$ hours of observation time with the uGMRT at redshift $7.1$ and $8.3$, respectively, when the mean neutral hydrogen fraction outside the targeted bubble is $ \sim 0.9$. The SKA1-Low should be able to detect these bubbles with more than $8 \sigma$ significance using only $\sim 100$ hrs of observations. The total observing time increases both for the uGMRT and SKA1-Low when the mean neutral hydrogen fraction outside the targeted bubble decreases. Further, we investigate the impact of foreground subtraction on the detectability and find the signal-to-noise ratio decreases when smaller bandwidth is used. More importantly, we show that the matched filtering method can measure ionized bubble radius and constrain HI-neutral fraction reasonably well, providing deeper insights into the source properties and the intergalactic medium., Comment: 31 pages, 19 figures, 1 table

Embeddings produced by pre-trained deep neural networks (DNNs) are widely used; however, their efficacy for downstream tasks can vary widely. We study the factors influencing transferability and out-of-distribution (OOD) generalization of pre-trained DNN embeddings through the lens of the tunnel effect hypothesis, which is closely related to intermediate neural collapse. This hypothesis suggests that deeper DNN layers compress representations and hinder OOD generalization. Contrary to earlier work, our experiments show this is not a universal phenomenon. We comprehensively investigate the impact of DNN architecture, training data, image resolution, and augmentations on transferability. We identify that training with high-resolution datasets containing many classes greatly reduces representation compression and improves transferability. Our results emphasize the danger of generalizing findings from toy datasets to broader contexts., Comment: Accepted to NeurIPS 2024

Foundation models in computational pathology promise to unlock the development of new clinical decision support systems and models for precision medicine. However, there is a mismatch between most clinical analysis, which is defined at the level of one or more whole slide images, and foundation models to date, which process the thousands of image tiles contained in a whole slide image separately. The requirement to train a network to aggregate information across a large number of tiles in multiple whole slide images limits these models' impact. In this work, we present a slide-level foundation model for H&E-stained histopathology, PRISM, that builds on Virchow tile embeddings and leverages clinical report text for pre-training. Using the tile embeddings, PRISM produces slide-level embeddings with the ability to generate clinical reports, resulting in several modes of use. Using text prompts, PRISM achieves zero-shot cancer detection and sub-typing performance approaching and surpassing that of a supervised aggregator model. Using the slide embeddings with linear classifiers, PRISM surpasses supervised aggregator models. Furthermore, we demonstrate that fine-tuning of the PRISM slide encoder yields label-efficient training for biomarker prediction, a task that typically suffers from low availability of training data; an aggregator initialized with PRISM and trained on as little as 10% of the training data can outperform a supervised baseline that uses all of the data.

The greatest demand for today's computing is machine learning. This paper analyzes three machine learning algorithms: transformers, spatial convolution, and FFT. The analysis is novel in three aspects. First, it measures the cost of memory access on an abstract memory hierarchy, instead of traditional time or space complexity. Second, the analysis is asymptotic and identifies the primary sources of the memory cost. Finally, the result is symbolic, which can be used to select algorithmic parameters such as the group size in grouped query attention for any dimension size and number of heads and the batch size for batched convolution for any image size and kernel size.

We propose a novel VQA dataset, BloomVQA, to facilitate comprehensive evaluation of large vision-language models on comprehension tasks. Unlike current benchmarks that often focus on fact-based memorization and simple reasoning tasks without theoretical grounding, we collect multiple-choice samples based on picture stories that reflect different levels of comprehension, as laid out in Bloom's Taxonomy, a classic framework for learning assessment widely adopted in education research. Our data maps to a novel hierarchical graph representation which enables automatic data augmentation and novel measures characterizing model consistency. We perform graded evaluation and reliability analysis on recent multi-modal models. In comparison to low-level tasks, we observe decreased performance on tasks requiring advanced comprehension and cognitive skills with up to 38.0\% drop in VQA accuracy. In comparison to earlier models, GPT-4V demonstrates improved accuracy over all comprehension levels and shows a tendency of bypassing visual inputs especially for higher-level tasks. Current models also show consistency patterns misaligned with human comprehension in various scenarios, demonstrating the need for improvement based on theoretically-grounded criteria., Comment: Accepted by ACL Findings (2024). Dataset available at https://huggingface.co/datasets/ygong/BloomVQA

It is believed that the first star-forming galaxies are the main drivers of cosmic reionization. It is usually assumed that there is a one-to-one relationship between the star formation rate (SFR) inside a galaxy and the host halo mass in semi-analytical/numerical modeling of large-scale reionization. However, more accurate simulations and observations suggest that the SFR and ionizing luminosity in galaxies may vary considerably even if the host halo mass is the same. This astrophysical scatter can introduce an additional non-Gaussianity in the HI 21-cm signal, which the power spectrum might not capture adequately. In this work, we have studied the impact of the scatter on the HI 21-cm bispectrum using semi-numerical simulations. Although the fractional change in the HI 21-cm bispectra due to the scatter is found to be more than a factor of $10$ at large scales ($k_1 \lesssim 1\, {\rm Mpc}^{-1}$) for $z=7.4$, it is found to be statistically insignificant. However, at small scales ($k_1\sim2.55$~Mpc$^{-1}$), we have found the impact due to the scatter to be high in magnitude ($|\langle \Delta B \rangle/B_{\text{no-scatter}}| \sim 1$) and statistically significant ($|\langle\Delta B\rangle/\sigma_{\Delta B}| \gtrsim 5$) at neutral fraction, $\overline{x}_{\rm HI}\sim 0.8$ for $z=7.4$. The impact due to scatter is found to be even more prominent ($|\langle \Delta B \rangle/B_{\text{no-scatter}}| \gtrsim 10$) at small scales for $z=10$ and $\overline{x}_{\rm HI}\sim 0.8$, but with reduced statistical significance ($|\langle\Delta B\rangle/\sigma_{\Delta B}| \sim 3$), compared to $z=7.4$ at the same neutral fraction. In the most optimistic scenario, SKA1-Low might be able to detect these signatures of astrophysical scatter, at $\sim 3\sigma$ and $\sim 5\sigma$ detection significance for $\overline{x}_{\rm HI} \sim$ 0.8 and 0.9 respectively, for the equilateral HI 21-cm bispectrum at $z=7.4$., Comment: 26 pages, 12 figures, comments are welcome, accepted in JCAP

Managing access between large numbers of distributed medical devices has become a crucial aspect of modern healthcare systems, enabling the establishment of smart hospitals and telehealth infrastructure. However, as telehealth technology continues to evolve and Internet of Things (IoT) devices become more widely used, they are also becoming increasingly exposed to various types of vulnerabilities and medical errors. In healthcare information systems, about 90\% of vulnerabilities emerged from misuse cases and human errors. As a result, there is a need for additional research and development of security tools to prevent such attacks. This article proposes a zero-trust-based context-aware framework for managing access to the main components of the cloud ecosystem, including users, devices and output data. The main goal and benefit of the proposed framework is to build a scoring system to prevent or alleviate misuse cases while using distributed medical devices in cloud-based healthcare information systems. The framework has two main scoring schemas to maintain the chain of trust. First, it proposes a critical trust score based on cloud-native micro-services of authentication, encryption, logging, and authorizations. Second, creating a bond trust scoring to assess the real-time semantic and syntactic analysis of attributes stored in a healthcare information system. The analysis is based on a pre-trained machine learning model to generate the semantic and syntactic scores. The framework also takes into account regulatory compliance and user consent to create a scoring system. The advantage of this method is that it is applicable to any language and adapts to all attributes as it relies on a language model, not just a set of predefined and limited attributes. The results show a high F1 score of 93.5%, which proves that it is valid for detecting misuse cases.

Continual learning is a subfield of machine learning, which aims to allow machine learning models to continuously learn on new data, by accumulating knowledge without forgetting what was learned in the past. In this work, we take a step back, and ask: "Why should one care about continual learning in the first place?". We set the stage by examining recent continual learning papers published at four major machine learning conferences, and show that memory-constrained settings dominate the field. Then, we discuss five open problems in machine learning, and even though they might seem unrelated to continual learning at first sight, we show that continual learning will inevitably be part of their solution. These problems are model editing, personalization and specialization, on-device learning, faster (re-)training and reinforcement learning. Finally, by comparing the desiderata from these unsolved problems and the current assumptions in continual learning, we highlight and discuss four future directions for continual learning research. We hope that this work offers an interesting perspective on the future of continual learning, while displaying its potential value and the paths we have to pursue in order to make it successful. This work is the result of the many discussions the authors had at the Dagstuhl seminar on Deep Continual Learning, in March 2023.

Many attempts have been made at estimating discrete emotions (calmness, anxiety, boredom, surprise, anger) and continuous emotional measures commonly used in psychology, namely `valence' (The pleasantness of the emotion being displayed) and `arousal' (The intensity of the emotion being displayed). Existing methods to estimate arousal and valence rely on learning from data sets, where an expert annotator labels every image frame. Access to an expert annotator is not always possible, and the annotation can also be tedious. Hence it is more practical to obtain self-reported arousal and valence values directly from the human in a real-time Human-Robot collaborative setting. Hence this paper provides an emotion data set (HRI-AVC) obtained while conducting a human-robot interaction (HRI) task. The self-reported pair of labels in this data set is associated with a set of image frames. This paper also proposes a spatial and temporal attention-based network to estimate arousal and valence from this set of image frames. The results show that an attention-based network can estimate valence and arousal on the HRI-AVC data set even when Arousal and Valence values are unavailable per frame.

Breast Cancer is the most common cancer among women, which is also visible in men, and accounts for more than 1 in 10 new cancer diagnoses each year. It is also the second most common cause of women who die from cancer. Hence, it necessitates early detection and tailored treatment. Early detection can provide appropriate and patient-based therapeutic schedules. Moreover, early detection can also provide the type of cyst. This paper employs class-level data augmentation, addressing the undersampled classes and raising their detection rate. This approach suggests two key components: class-level data augmentation on structure-preserving stain normalization techniques to hematoxylin and eosin-stained images and transformer-based ViTNet architecture via transfer learning for multiclass classification of breast cancer images. This merger enables categorizing breast cancer images with advanced image processing and deep learning as either benign or as one of four distinct malignant subtypes by focusing on class-level augmentation and catering to unique characteristics of each class with increasing precision of classification on undersampled classes, which leads to lower mortality rates associated with breast cancer. The paper aims to ease the duties of the medical specialist by operating multiclass classification and categorizing the image into benign or one of four different malignant types of breast cancers.

The use of artificial intelligence to enable precision medicine and decision support systems through the analysis of pathology images has the potential to revolutionize the diagnosis and treatment of cancer. Such applications will depend on models' abilities to capture the diverse patterns observed in pathology images. To address this challenge, we present Virchow, a foundation model for computational pathology. Using self-supervised learning empowered by the DINOv2 algorithm, Virchow is a vision transformer model with 632 million parameters trained on 1.5 million hematoxylin and eosin stained whole slide images from diverse tissue and specimen types, which is orders of magnitude more data than previous works. The Virchow model enables the development of a pan-cancer detection system with 0.949 overall specimen-level AUC across 17 different cancer types, while also achieving 0.937 AUC on 7 rare cancer types. The Virchow model sets the state-of-the-art on the internal and external image tile level benchmarks and slide level biomarker prediction tasks. The gains in performance highlight the importance of training on massive pathology image datasets, suggesting scaling up the data and network architecture can improve the accuracy for many high-impact computational pathology applications where limited amounts of training data are available.

Continual learning (CL) in deep neural networks (DNNs) involves incrementally accumulating knowledge in a DNN from a growing data stream. A major challenge in CL is that non-stationary data streams cause catastrophic forgetting of previously learned abilities. A popular solution is rehearsal: storing past observations in a buffer and then sampling the buffer to update the DNN. Uniform sampling in a class-balanced manner is highly effective, and better sample selection policies have been elusive. Here, we propose a new sample selection policy called GRASP that selects the most prototypical (easy) samples first and then gradually selects less prototypical (harder) examples. GRASP has little additional compute or memory overhead compared to uniform selection, enabling it to scale to large datasets. Compared to 17 other rehearsal policies, GRASP achieves higher accuracy in CL experiments on ImageNet. Compared to uniform balanced sampling, GRASP achieves the same performance with 40% fewer updates. We also show that GRASP is effective for CL on five text classification datasets., Comment: Accepted to the Conference on Lifelong Learning Agents (CoLLAs) 2024

Image Augmentations are widely used to reduce overfitting in neural networks. However, the explainability of their benefits largely remains a mystery. We study which layers of residual neural networks (ResNets) are most affected by augmentations using Centered Kernel Alignment (CKA). We do so by analyzing models of varying widths and depths, as well as whether their weights are initialized randomly or through transfer learning. We find that the pattern of how the layers are affected depends on the model's depth, and that networks trained with augmentation that use information from two images affect the learned weights significantly more than augmentations that operate on a single image. Deeper layers of ResNets initialized with ImageNet-1K weights and fine-tuned receive more impact from the augmentations than early layers. Understanding the effects of image augmentations on CNNs will have a variety of applications, such as determining how far back one needs to fine-tune a network and which layers should be frozen when implementing layer freezing algorithms.

A key aim of this paper is to explore how our professional tasks as geoscientists and petroleum engineers can be completed more effectively making use of tools powered by Artificial Intelligence (AI), offered in commercial platforms now readily available to individual users. This paper intends to provide some guidance, but at the same time does not claim to be comprehensive or conclusive in any way. The paper presents a utility assessment from the research and teaching vantage points of two professors and one student, from geosciences and petroleum engineering departments. After a brief overview of the new technologies, some key questions raised include: How can one assess originality of class papers by students and research papers by their professors? How will the contribution of intelligent devices be acknowledged? Will the presentation of conference papers by author avatars be accepted by the organizing committee?

Pre-trained deep neural networks (DNNs) are being widely deployed by industry for making business decisions and to serve users; however, a major problem is model decay, where the DNN's predictions become more erroneous over time, resulting in revenue loss or unhappy users. To mitigate model decay, DNNs are retrained from scratch using old and new data. This is computationally expensive, so retraining happens only once performance significantly decreases. Here, we study how continual learning (CL) could potentially overcome model decay in large pre-trained DNNs and greatly reduce computational costs for keeping DNNs up-to-date. We identify the "stability gap" as a major obstacle in our setting. The stability gap refers to a phenomenon where learning new data causes large drops in performance for past tasks before CL mitigation methods eventually compensate for this drop. We test two hypotheses to investigate the factors influencing the stability gap and identify a method that vastly reduces this gap. In large-scale experiments for both easy and hard CL distributions (e.g., class incremental learning), we demonstrate that our method reduces the stability gap and greatly increases computational efficiency. Our work aligns CL with the goals of the production setting, where CL is needed for many applications., Comment: Accepted to TMLR 2024

Assessing the artness of AI-generated images continues to be a challenge within the realm of image generation. Most existing metrics cannot be used to perform instance-level and reference-free artness evaluation. This paper presents ArtScore, a metric designed to evaluate the degree to which an image resembles authentic artworks by artists (or conversely photographs), thereby offering a novel approach to artness assessment. We first blend pre-trained models for photo and artwork generation, resulting in a series of mixed models. Subsequently, we utilize these mixed models to generate images exhibiting varying degrees of artness with pseudo-annotations. Each photorealistic image has a corresponding artistic counterpart and a series of interpolated images that range from realistic to artistic. This dataset is then employed to train a neural network that learns to estimate quantized artness levels of arbitrary images. Extensive experiments reveal that the artness levels predicted by ArtScore align more closely with human artistic evaluation than existing evaluation metrics, such as Gram loss and ArtFID., Comment: Published in IEEE Transactions on Multimedia