Search

Foundation models are rapidly being developed for computational pathology applications. However, it remains an open question which factors are most important for downstream performance with data scale and diversity, model size, and training algorithm all playing a role. In this work, we propose algorithmic modifications, tailored for pathology, and we present the result of scaling both data and model size, surpassing previous studies in both dimensions. We introduce two new models: Virchow2, a 632 million parameter vision transformer, and Virchow2G, a 1.9 billion parameter vision transformer, each trained with 3.1 million histopathology whole slide images, with diverse tissues, originating institutions, and stains. We achieve state of the art performance on 12 tile-level tasks, as compared to the top performing competing models. Our results suggest that data diversity and domain-specific methods can outperform models that only scale in the number of parameters, but, on average, performance benefits from the combination of domain-specific methods, data scale, and model scale.

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.

Self-supervised learning (SSL) has emerged as a key technique for training networks that can generalize well to diverse tasks without task-specific supervision. This property makes SSL desirable for computational pathology, the study of digitized images of tissues, as there are many target applications and often limited labeled training samples. However, SSL algorithms and models have been primarily developed in the field of natural images and whether their performance can be improved by adaptation to particular domains remains an open question. In this work, we present an investigation of modifications to SSL for pathology data, specifically focusing on the DINOv2 algorithm. We propose alternative augmentations, regularization functions, and position encodings motivated by the characteristics of pathology images. We evaluate the impact of these changes on several benchmarks to demonstrate the value of tailored approaches., Comment: Presented at DCA in MI Workshop, CVPR 2024

Large Language Models (LLMs) have demonstrated remarkable proficiency in understanding and generating natural language. However, their capabilities wane in highly specialized domains underrepresented in the pretraining corpus, such as physical and biomedical sciences. This work explores how to repurpose general LLMs into effective task solvers for specialized domains. We introduce a novel, model-agnostic framework for learning custom input tags, which are parameterized as continuous vectors appended to the LLM's embedding layer, to condition the LLM. We design two types of input tags: domain tags are used to delimit specialized representations (e.g., chemical formulas) and provide domain-relevant context; function tags are used to represent specific functions (e.g., predicting molecular properties) and compress function-solving instructions. We develop a three-stage protocol to learn these tags using auxiliary data and domain knowledge. By explicitly disentangling task domains from task functions, our method enables zero-shot generalization to unseen problems through diverse combinations of the input tags. It also boosts LLM's performance in various specialized domains, such as predicting protein or chemical properties and modeling drug-target interactions, outperforming expert models tailored to these tasks., Comment: ICML 2024

In certain types of cancerous tissue, mitotic count has been shown to be associated with tumor proliferation, poor prognosis, and therapeutic resistance. Due to the high inter-rater variability of mitotic counting by pathologists, convolutional neural networks (CNNs) have been employed to reduce the subjectivity of mitosis detection in hematoxylin and eosin (H&E)-stained whole slide images. However, most existing models have performance that lags behind expert panel review and only incorporate visual information. In this work, we demonstrate that pre-trained large-scale vision-language models that leverage both visual features and natural language improve mitosis detection accuracy. We formulate the mitosis detection task as an image captioning task and a visual question answering (VQA) task by including metadata such as tumor and scanner types as context. The effectiveness of our pipeline is demonstrated via comparison with various baseline models using 9,501 mitotic figures and 11,051 hard negatives (non-mitotic figures that are difficult to characterize) from the publicly available Mitosis Domain Generalization Challenge (MIDOG22) dataset., Comment: Submitted to IEEE ISBI 2024. Under review

This paper investigates the application of quantum computing technology to airline gate-scheduling quadratic assignment problems (QAP). We explore the quantum computing hardware architecture and software environment required for porting classical versions of these type of problems to quantum computers. We discuss the variational quantum eigensolver and the inclusion of space-efficient graph coloring to the Quadratic Unconstrained Binary Optimization (QUBO). These enhanced quantum computing algorithms are tested with an 8 gate and 24 flight test case using both the IBM quantum computing simulator and a 27 qubit superconducting transmon IBM quantum computing hardware platform.

We present a characterization of the completeness of the field of real numbers in the form of a \emph{collection of several equivalent statements} borrowed from algebra, real analysis, general topology, and non-standard analysis. We also discuss the completeness of non-Archimedean fields and present several examples of such fields. As an application, we exploit the characterization of the reals to argue that the Leibniz-Euler infinitesimal calculus in the $17^\textrm{th}$-$18^\textrm{th}$ centuries was already a rigorous branch of mathematics -- at least much more rigorous than most contemporary mathematicians prefer to believe. By advocating our particular historical point of view, we hope to provoke a discussion on the importance of mathematical rigor in mathematics and science in general. This article is directed to all mathematicians and scientists who are interested in the foundations of mathematics and the history of infinitesimal calculus., Comment: 29 pages. arXiv admin note: substantial text overlap with arXiv:1109.2098

Determining phenotype from genetic data is a fundamental challenge. Influenza A viruses undergo rapid antigenic drift and identification of emerging antigenic variants is critical to the vaccine selection process. Using former seasonal influenza A(H1N1) viruses, hemagglutinin sequence and corresponding antigenic data were analyzed in combination with 3-D structural information. We attributed variation in hemagglutination inhibition to individual amino acid substitutions and quantified their antigenic impact, validating a subset experimentally using reverse genetics. Substitutions identified as low-impact were shown to be a critical component of influenza antigenic evolution and by including these, as well as the high-impact substitutions often focused on, the accuracy of predicting antigenic phenotypes of emerging viruses from genotype was doubled. The ability to quantify the phenotypic impact of specific amino acid substitutions should help refine techniques that predict the fitness and evolutionary success of variant viruses, leading to stronger theoretical foundations for selection of candidate vaccine viruses.

We present a new class of high-order variational integrators on Lie groups. We show that these integrators are symplectic, momentum preserving, and can be constructed to be of arbitrarily high-order, or can be made to converge geometrically. Furthermore, these methods are stable and accurate for very large time steps. We demonstrate the construction of one such variational integrator for the rigid body, and discuss how this construction could be generalized to other related Lie group problems. We close with several numerical examples which demonstrate our claims, and discuss further extensions of our work.

In this paper, we present a new variational integrator for problems in Lagrangian mechanics. Using techniques from Galerkin variational integrators, we construct a scheme for numerical integration that converges geometrically, and is symplectic and momentum preserving. Furthermore, we prove that under appropriate assumptions, variational integrators constructed using Galerkin techniques will yield numerical methods that are in a certain sense optimal, converging at the same rate as the best possible approximation in a certain function space. We further prove that certain geometric invariants also converge at an optimal rate, and that the error associated with these geometric invariants is independent of the number of steps taken. We close with several numerical examples that demonstrate the predicted rates of convergence., Comment: 37 pages, 15 figures

We present a characterization of the completeness of the field of real numbers in the form of a \emph{collection of ten equivalent statements} borrowed from algebra, real analysis, general topology and non-standard analysis. We also discuss the completeness of non-Archimedean fields and present several examples of such fields. As an application we exploit one of our results to argue that the Leibniz infinitesimal calculus in the $18^\textrm{th}$ century was already a rigorous branch of mathematics -- at least much more rigorous than most contemporary mathematicians prefer to believe. By advocating our particular historical point of view, we hope to provoke a discussion on the importance of mathematical rigor in mathematics and science in general. We believe that our article will be of interest for those readers who teach courses on abstract algebra, real analysis, general topology, logic and the history of mathematics., Comment: 32

The main goal of this project is to prove the equivalency of several characterizations of completeness of Archimedean ordered fields; some of which appear in most modern literature as theorems following from the Dedekind completeness of the real numbers, while a couple are not as well known and have to do with other areas of mathematics, such as nonstandard analysis. Continuing, we study the completeness of non-Archimedean fields, and provide several examples of such fields with varying degrees of properties, using nonstandard analysis to produce some relatively "nice" (in particular, they are Cantor complete) final examples. As a small detour, we present a short construction of the real numbers using methods from nonstandard analysis.

We describe the Dedekind cuts explicitly in terms of non-standard rational numbers. This leads to another construction of a Dedekind complete totally ordered field or, equivalently, to another proof of the consistency of the axioms of the real numbers. We believe that our construction is simpler and shorter than the classical Dedekind construction and Cantor construction of such fields assuming some basic familiarity with non-standard analysis., Comment: 5 pages

We have conducted tests at the Fermilab Booster of ionization chambers to be used as monitors of the NuMI neutrino beamline. The chambers were exposed to proton fluxes of up to 10$^{12}$ particles/cm$^2$/1.56$\mu$s. We studied space charge effects which can reduce signal collection from the chambers at large charged particle beam intensities., Comment: submitted to IEEE Trans Nucl. Sci

A method and system of detecting, localizing, and characterizing a defect at one or more spatial points of interest on a structure. The method may include collecting first data in a first state using one or more transducers on the structure, collecting second data in a second state subsequent to the first state, computing a scattered impulse response based on the collected first data and the collected second data, comparing the scattered impulse response with an estimated scattered impulse response corresponding to the case when damage is present at one or more spatial points of interest on the structure, and combining the generated comparison results to detect, localize, and characterize a defect at the one or more spatial points of interest on the structure.

An external degree is one that may be earned outside the central structure of a college or university. The most common form of external degree program is the evening college in which students, who are usually working adults, meet the standard requirements for a four-year degree by taking a part-time program after work. No allowances are made in evening college programs for what adult students may have learned outside the classroom, even though this learning may be equivalent to what they are required to learn for their degrees. Sensing this problem, a number of programs were developed in the late forties and fifties to take advantage of the rich experience that many adults have had in their lives. Most often such programs provide not only an opportunity for credit for life experience but also for teaching methods and requirements that seem to be more appropriate for mature students. The adult degree programs of nine schools are examined in the report; Brooklyn College, Queens College; Mundelein College; Roosevelt University; University of Oklahoma; Syracuse University; Goddard College; University of South Florida; State University of New York College, Brockport. The most significant findings of the study are: (1) these programs seem to be reacting to the move toward vocational orientation in undergraduate programs. (2) The programs are moving away from their unique designs toward compatibility with more traditional programs. (3) Lack of full time faculties keeps the staff of these programs from exercising authority within their institutions. (Author/PG)

Hall, Leidecker, and DiMarco write that upward performance appraisals are supposed to improve communication and increase worker satisfaction. Concerns are related to subordinate qualifications and bias, lack of time, and acceptance of results. Three types are subordinate oriented, cross-hierarchical, and self-inclusive. Stoner's reaction argues that Hall et al. do not define terms or suggest a new taxonomy. (SK)

In 1970, the Illinois Legislature passed bills requiring that elementary school districts institute procedures to identify and treat preschool children who, without such intervention, would not progress satisfactorily in a normal school environment. Efforts in this regard resulted in the screening instrument called Developmental Indicators for the Assessment of Learning (DIAL). Despite initial administrations of the battery and several research efforts, information regarding the reliability and predictive validity of the DIAL was insufficient to promote its general use for preschool screening. This project was funded in an effort to supply the missing data and to address additional questions. The major activities and findings of the project are described in the second section of the report. These concern the content validity of the DIAL as judged by consulting experts in child behavior and development, the predictive validity of the DIAL, the reliability of DIAL administration and scoring procedures, and various recommended modifications of the DIAL. The final section of the report summarizes conclusions and outlines recommendations concerning subsequent steps in the development and use of the DIAL. (RC)

Distance educators need to forge new partnerships with the traditional university and collaborate in creating new systems, academic course software, and supporting networks for information transmission. (Author/AEF)

Through a combination of prior coursework, contract learning, credit for experience, standardized tests, and other methods of evaluating prior learning, the Empire State College (New York) allows adults to complete undergraduate degree requirements. The nonresidential institution serves students throughout the state, with faculty in over 40 regional learning locations. (Author/MSE)

In 2005, the Solar System Exploration Strategic Roadmap Conmrittee (chartered by NASA to develop the roadmap for Solar System Exploration Missions for the coming decades) found itself posed with the difficult problem of sorting through several mission concepts and determining their relative costs. While detailed mission studies are the normal approach to costing, neither the budget nor schedule allotted to the conmrittee could support such studies. Members of the Jet Propulsion Laboratory (JPL) supporting the conmrittee were given the challenge of developing a semi-quantitative approach that could provide the relative costs of these missions, without requiring an in depth study of the missions. In response to this challenge, a rapid cost assessment methodology based on a set of mission cost/complexity indexes was developed. This methodology also underwent two separate validations, one comparing its results when applied to historical missions, and another comparing its estimates against those of veteran space mission managers. Remarkably good agreement was achieved, suggesting that this approach provides an effective early indication of space mission costs.

This slide presentation reviews the development of a rapid cost assessment models for evaluation of exploration missions through the application of complexity based cost indices. In Fall of 2004, NASA began developing 13 documents, known as "strategic roadmaps," intended to outline a strategy for space exploration over the next 30 years. The Third Strategic Roadmap, The Strategic Roadmap for Solar System Exploration, focused on strategy for robotic exploration of the Solar System. Development of the Strategic Roadmap for Solar System Exploration led to the investigation of a large variety of missions. However, the necessity of planning around scientific inquiry and budgetary constraints made it necessary for the roadmap development team to evaluate potential missions not only for scientific return but also cost. Performing detailed cost studies for each of the large number of missions was impractical given the time constraints involved and lack of detailed mission studies; so a method of rapid cost assessment was developed by us to allow preliminary analysis. It has been noted that there is a strong correlation between complexity and cost and schedule of planetary missions. While these correlations were made after missions had been built and flown (successfully or otherwise), it seemed likely that a similar approach could provide at least some relative cost ranking. Cost estimation relationships (CERs) have been developed based on subsystem design choices. These CERs required more detailed information than available, forcing the team to adopt a more high level approach. Costing by analogy has been developed for small satellites, however, planetary exploration missions provide such varying spacecraft requirements that there is a lack of adequately comparable missions that can be used for analogy.

A suite of computer programs has been developed for special test equipment (STE) that is used in verification testing of the Health Management Computer Integrated Rack Assembly (HMCIRA), a ground-based system of analog and digital electronic hardware and software for "flight-like" testing for development of components of an advanced health-management system for the space shuttle main engine (SSME). The STE software enables the STE to simulate the analog input and the data flow of an SSME test firing from start to finish.

Cover glasses have been a necessary and integral part of space solar arrays since their inception. The main function of the cover glass is to protect the underlying solar cell from the harsh radiation environment of space. They are formed either from fused silica or specially formulated ceria doped glass types that are resistant to radiation damage, for example Pilkington's CMX, CMG, CMO. Solar cells have steadily increased in performance over the past years, from Silicon cells through textured Silicon cells to GaAs cells and the multijunction cells of today. The optimum coverglass solution for each of these cells has been different. The glass itself has also evolved. In some cases it has had its expansion coefficient matched to the cell substrate material, and in addition, added value has been derived from the application of thin film optical coatings to the coverglass. In the majority of cases this has taken the form of a single layer of MgF2 which acts as an antireflection coating. There are also conductive coatings to address electrostatic discharge issues (ESD) and Ultra Violet Reflective (UVR) and Infrared Reflective (IRR) coatings designed for thermal enhancement. Each type of coating can be applied singly or in combination. This paper describes a new type of UVR/IRR (or blue red reflector BRR) specifically designed for triple junction solar cells. For space applications, where radiation is the principal mechanism for removing heat from the satellite, it is the emittance and solar absorptance that primarily determine the temperature of the array. It is therefore essential that any coatings designed to have an effect on the temperature by reducing the solar absorption have a minimal effect on the overall emittance.

The MARTE (Mars Astrobiology Research and Technology Experiment) project, an ASTEP field experiment, is exploring for a hypothesized subsurface anaerobic chemoautotrophic biosphere in the region of the Tinto River- or Rio Tinto- in southwestern Spain. It is also demonstrating technology needed to search for a subsurface biosphere on Mars. The project has three primary objectives: (1) search for and characterize subsurface life at Rio Tinto along with the physical and chemical properties and sustaining energy sources of its environment, (2) perform a high fidelity simulation of a robotic Mars drilling mission to search for life, and (3) demonstrate the drilling, sample handling, and instrument technologies relevant to searching for life on Mars. The simulation of the robotic drilling mission is guided by the results of the aseptic drilling campaign to search for life at Rio Tinto. This paper describes results of the first phase of the aseptic drilling campaign.

The keyword method, originally developed as an instructional technique in foreign language acquisition, was tested for its application in other school learning tasks and as a self-initiated study strategy. Two classes containing a total of 40 eighth grade students served as the experimental and control groups. The tasks were learning to pair technical terms with their definitions and learning to pair proper surnames with facts, events, and accomplishments. After five training sessions the experimental group students were able to carry out the sequence of operations constituting the keyword method, including the generation of concrete keywords and of compound images that bridge keywords and response terms. The experimental group performed significantly better than the control group, exceeding the control group by 40% on the posttest and 36% on the other posttest. Subsequent unprompted tests showed substantially higher performance by students trained in the keyword method than by those students from the control group. (Author/RL)

After 44 seventh grade students were matched in pairs according to their reading achievement test scores, one member of each pair received training in a cross-classification strategy (matrix outlining), while the other member of each pair received unrelated training. Cross-classifiable information is information that can be categorized along two dimensions simultaneously, such as countries classified in name categories and attribute (type of economy) categories. The trained students had significantly higher scores than untrained students on their free recall of 36 disorganized sentences and on their compare-and-contrast essays. Both tests were given without strategy prompts. The posttest data also indicated that the training was especially effective in increasing learning by low achieving students. (RL)