Search

This paper details the efforts of the WisPerMed team in the BioLaySumm2024 Shared Task on automatic lay summarization in the biomedical domain, aimed at making scientific publications accessible to non-specialists. Large language models (LLMs), specifically the BioMistral and Llama3 models, were fine-tuned and employed to create lay summaries from complex scientific texts. The summarization performance was enhanced through various approaches, including instruction tuning, few-shot learning, and prompt variations tailored to incorporate specific context information. The experiments demonstrated that fine-tuning generally led to the best performance across most evaluated metrics. Few-shot learning notably improved the models' ability to generate relevant and factually accurate texts, particularly when using a well-crafted prompt. Additionally, a Dynamic Expert Selection (DES) mechanism to optimize the selection of text outputs based on readability and factuality metrics was developed. Out of 54 participants, the WisPerMed team reached the 4th place, measured by readability, factuality, and relevance. Determined by the overall score, our approach improved upon the baseline by approx. 5.5 percentage points and was only approx 1.5 percentage points behind the first place., Comment: 4 pages, 6 figure, 3 tables, submitted to: BIONLP 2024 and Shared Tasks @ ACL 2024

Automated medical image analysis systems often require large amounts of training data with high quality labels, which are difficult and time consuming to generate. This paper introduces Radiology Object in COntext version 2 (ROCOv2), a multimodal dataset consisting of radiological images and associated medical concepts and captions extracted from the PMC Open Access subset. It is an updated version of the ROCO dataset published in 2018, and adds 35,705 new images added to PMC since 2018. It further provides manually curated concepts for imaging modalities with additional anatomical and directional concepts for X-rays. The dataset consists of 79,789 images and has been used, with minor modifications, in the concept detection and caption prediction tasks of ImageCLEFmedical Caption 2023. The dataset is suitable for training image annotation models based on image-caption pairs, or for multi-label image classification using Unified Medical Language System (UMLS) concepts provided with each image. In addition, it can serve for pre-training of medical domain models, and evaluation of deep learning models for multi-task learning., Comment: Accepted for Scientific Data

Recent advances in natural language processing (NLP) can be largely attributed to the advent of pre-trained language models such as BERT and RoBERTa. While these models demonstrate remarkable performance on general datasets, they can struggle in specialized domains such as medicine, where unique domain-specific terminologies, domain-specific abbreviations, and varying document structures are common. This paper explores strategies for adapting these models to domain-specific requirements, primarily through continuous pre-training on domain-specific data. We pre-trained several German medical language models on 2.4B tokens derived from translated public English medical data and 3B tokens of German clinical data. The resulting models were evaluated on various German downstream tasks, including named entity recognition (NER), multi-label classification, and extractive question answering. Our results suggest that models augmented by clinical and translation-based pre-training typically outperform general domain models in medical contexts. We conclude that continuous pre-training has demonstrated the ability to match or even exceed the performance of clinical models trained from scratch. Furthermore, pre-training on clinical data or leveraging translated texts have proven to be reliable methods for domain adaptation in medical NLP tasks., Comment: Accepted at LREC-COLING 2024

We calculate the intersection ring of three-dimensional graph manifolds with rational coefficients and give an algebraic characterization of these rings when the manifold's underlying graph is a tree. We are able to use this characterization to show that the intersection ring obstructs arbitrary three-manifolds from being homology cobordant to certain graph manifolds., Comment: 15 pages, 9 figures, addressed referee comments

Cochran defined the nth-order integral Alexander module of a knot in the three sphere as the first homology group of the knot's (n+1)th-iterated abelian cover. The case n=0 gives the classical Alexander module (and polynomial). After a localization, one can get a finitely presented module over a principal ideal domain, from which one can extract a higher-order Alexander polynomial. We present an algorithm to compute the first-order Alexander module for any knot. As applications, we show that these higher-order Alexander polynomials provide a better bound on the knot genus than does the classical Alexander polynomial, and that they detect mutation. Included in this algorithm is a solution to the word problem in finitely presented Z[Z]-modules., Comment: 20 pages, 3 figures, added some references, added result on mutation

Let T denote the group of smooth concordance classes of topologically sice knots. We show that the first quotient in the bipolar filtration of T (i.e. 0-bipolar knots modulo 1-bipolar knots) has infinite rank, even modulo Alexander polynomial one knots. Any 0-bipolar knot has vanishing tau-, epsilon-, and s-invariants. We prove the result using d-invariants associated to the 2-fold branched covers of knot complements., Comment: 9 pages, 5 figures

We propose and analyze a structure with which to organize the difference between a knot in the 3-sphere bounding a topologically embedded 2-disk in the 4-ball and it bounding a smoothly embedded disk. The n-solvable filtration of the topological knot concordance group, due to Cochran-Orr-Teichner, may be complete in the sense that any knot in the intersection of its terms may well be topologically slice. However, the natural extension of this filtration to what is called the n-solvable filtration of the smooth knot concordance group, is unsatisfactory because any topologically slice knot lies in every term of the filtration. To ameliorate this we investigate a new filtration, {B_n}, that is simultaneously a refinement of the n-solvable filtration and a generalization of notions of positivity studied by Gompf and Cochran. We show that each B_n/B_{n+1} has infinite rank. But our primary interest is in the induced filtration, {T_n}, on the subgroup, T, of knots that are topologically slice. We prove that T/T_0 is large, detected by gauge-theoretic invariants and the tau, s, and epsilon-invariants; while the non-triviliality of T_0/T_1 can be detected by certain d-invariants. All of these concordance obstructions vanish for knots in T_1. Nonetheless, going beyond this, our main result is that T_1/T_2 has positive rank. Moreover under a "weak homotopy-ribbon" condition, we show that each T_n/T_{n+1} has positive rank. These results suggest that, even among topologically slice knots, the fundamental group is responsible for a wide range of complexity., Comment: 41 pages, slightly revised introduction, minor corrections and up-dated references, this is the final version to appear in Geometry and Topology

We consider the question: "If the zero-framed surgeries on two oriented knots in the 3-sphere are integral homology cobordant, preserving the homology class of the positive meridians, are the knots themselves concordant?" We show that this question has a negative answer in the smooth category, even for topologically slice knots. To show this we first prove that the zero-framed surgery on K is Z-homology cobordant to the zero-framed surgery on many of its winding number one satellites P(K). Then we prove that in many cases the tau and s-invariants of K and P(K) differ. Consequently neither tau nor s is an invariant of the smooth homology cobordism class of the zero-framed surgery. We also show, that a natural rational version of this question has a negative answer in both the topological and smooth categories, by proving similar results for K and its (p,1)-cables., Comment: 15 pages, 8 figures

The following is a long-standing open question: "If the zero-framed surgeries on two knots in the 3-sphere are integral homology cobordant, are the knots themselves concordant?" We show that an obvious rational version of this question has a negative answer. Namely, we give examples of knots whose zero-framed surgeries are rational homology cobordant 3-manifolds, wherein the knots are not rationally concordant (that is not concordant in any rational homology S^3 x [0,1]). Specifically, we prove that, for any positive integer p and any knot K, the zero framed surgery on K is Z[1/p]-homology cobordant to the zero framed surgery on its (p,1) cable. Then we observe that most knots are not rationally concordant to their (p,1) cables., Comment: 9 pages, 6 figures; Reference added and 2 errors corrected in introduction

We define families of invariants for elements of the mapping class group of S, a compact orientable surface. Fix any characteristic subgroup H of pi_1(S) and restrict to J(H), any subgroup of mapping classes that induce the identity modulo H. To any unitary representation, r of pi_1(S)/H we associate a higher-order rho_r-invariant and a signature 2-cocycle sigma_r. These signature cocycles are shown to be generalizations of the Meyer cocycle. In particular each rho_r is a quasimorphism and each sigma_r is a bounded 2-cocycle on J(H). In one of the simplest non-trivial cases, by varying r, we exhibit infinite families of linearly independent quasimorphisms and signature cocycles. We show that the rho_r restrict to homomorphisms on certain interesting subgroups. Many of these invariants extend naturally to the full mapping class group and some extend to the monoid of homology cylinders based on S., Comment: 38 pages. This is final version for publication in IMRN, deleted some material and many references (sorry-at referee's insistence)

It is known that knot Floer homology detects the genus and Alexander polynomial of a knot. We investigate whether knot Floer homology of $K$ detects more structure of minimal genus Seifert surfaces for $K$. We define an invariant of algebraically slice, genus one knots and provide examples to show that knot Floer homology does not detect this invariant. Finally, we remark that certain metabelian $L^2$-signatures bound this invariant from below., Comment: 6 pages, 5 figures

For certain classes of knots we define geometric invariants called higher-order genera. Each of these invariants is a refinement of the slice genus of a knot. We find lower bounds for the higher-order genera in terms of certain von Neumann $\rho$-invariants, which we call higher-order signatures. The higher-order genera offer a refinement of the Grope filtration of the knot concordance group., Comment: 10 pages, 7 figures, added references, fixed typos, this version will be published in International Mathematics Research Notices

We consider the Grope filtration of the classical knot concordance group that was introduced in a paper of Cochran, Orr and Teichner. Our main result is that successive quotients at each stage in this filtration have infinite rank. We also establish the analogous result for the Grope filtration of the concordance group of string links consisting of more than one component., Comment: 16 pages, 14 figures

We introduce a geometric invariant of knots in the three-sphere, called the first-order genus, that is derived from certain 2-complexes called gropes, and we show it is computable for many examples. While computing this invariant, we draw some interesting conclusions about the structure of a general Seifert surface for some knots., Comment: 14 pages, 17 figures