Contrastive learning (CL) is a prevalent technique for training embedding models, which pulls semantically similar examples (positives) closer in the representation space while pushing dissimilar ones (negatives) further apart. A key source of negatives are "in-batch" examples, i.e., positives from other examples in the batch. Effectiveness of such models is hence strongly influenced by the size and quality of training batches. In this work, we propose *Breaking the Batch Barrier* (B3), a novel batch construction strategy designed to curate high-quality batches for CL. Our approach begins by using a pretrained teacher embedding model to rank all examples in the dataset, from which a sparse similarity graph is constructed. A community detection algorithm is then applied to this graph to identify clusters of examples that serve as strong negatives for one another. The clusters are then used to construct batches that are rich in in-batch negatives. Empirical results on the MMEB multimodal embedding benchmark (36 tasks) demonstrate that our method sets a new state of the art, outperforming previous best methods by +1.3 and +2.9 points at the 7B and 2B model scales, respectively. Notably, models trained with B3 surpass existing state-of-the-art results even with a batch size as small as 64, which is 4–16× smaller than that required by other methods. Moreover, experiments show that B3 generalizes well across domains and tasks, maintaining strong performance even when trained with considerably weaker teachers.

Data selection techniques, which adaptively select datapoints inside the training loop, have demonstrated empirical benefits in reducing the number of gradient steps to train neural models. However, these techniques have so far largely been applied to classification. In this work, we study their applicability to language model pretraining, a highly time-intensive task. We propose a simple modification to an existing data selection technique (reducible hold-out loss training) in order to adapt it to the sequence losses typical in language modeling. We experiment on both autoregressive and masked language modelling, and show that applying data selection to pretraining offers notable benefits including a 4.3% reduction in total number of steps, a 21.5% steps reduction in average, to an intermediate target perplexity, over the course of pretraining an autoregressive language model. Further, data selection trained language models demonstrate significantly better generalization ability on out of domain datasets - 7.9% reduction in total number of steps and 23.2% average steps reduction to an intermediate target perplexity.

Recent work has aimed to improve LLM generations by filtering out hallucinations, thereby improving the precision of the information in responses.Correctness of a long-form response, however, also depends on the recall of multiple pieces of information relevant to the question.In this paper, we introduce Atomic Self-Consistency (ASC), a technique for improving the recall of relevant information in an LLM response.ASC follows recent work, Universal Self-Consistency (USC) in using multiple stochastic samples from an LLM to improve the long-form response.Unlike USC which only focuses on selecting the best single generation, ASC picks authentic subparts from the samples and merges them into a superior composite answer.Through extensive experiments and ablations, we show that merging relevant subparts of multiple samples performs significantly better than picking a single sample.ASC demonstrates significant gains over USC on multiple factoids and open-ended QA datasets -ASQA, QAMPARI, QUEST, ELI5 with ChatGPT and Llama3.Our analysis also reveals untapped potential for enhancing long-form generations using approach of merging multiple samples.

In this paper, we tackle the problem of fake news detection from social media by exploiting the presence of echo chamber communities (communities sharing same beliefs) that exist within the social network of the users. By modeling the echo-chambers as closely-connected communities within the social network, we represent a news article as a 3-mode tensor of the structure -and propose a tensor factorization based method to encode the news article in a latent embedding space preserving the community structure. We also propose an extension of the above method, which jointly models the community and content information of the news article through a coupled matrix-tensor factorization framework. We empirically demonstrate the efficacy of our method for the task of Fake News Detection over two real-world datasets.

Condition-based monitoring (CBM), with its tremendous scope in improving operational cost efficiency, has become an important component in most industrial machine maintenance frameworks. While many strategies already exist for CBM, this paper presents a novel pattern analysis framework, which focuses on improving reliability of real-time fault detection. The introduced framework is based upon a novel feature selection method, which selects good, reliable, and consistent features. Graphical indices are proposed in this paper which try and quantify the goodness of features and accordingly rank them for the feature selection procedure. The presented algorithm also takes into account the possibility of corrupt data creeping in during data collection, and takes necessary steps to discard them. Additionally, novel methods for organizing training data and a method for sensitive position identification for placing sensor(s) have also been introduced, for further improving the quality of selected features. To validate the framework, experiments have been performed on an air compressor for real-time detection of leakage inlet valve fault and leakage outlet valve fault, and also on an induction motor for detecting presence of faulty bearings. The findings clearly show substantial improvement in fault detection performance and confirm the effectiveness of proposed framework.

I1 Proceedings of the Fifteenth Annual UT- KBRIN Bioinformatics Summit 2016 Eric C. Rouchka, Julia H. Chariker, Benjamin J. Harrison, Juw Won Park P1 CC-PROMISE: Projection onto the Most Interesting Statistical Evidence (PROMISE) with Canonical Correlation to integrate gene expression and methylation data with multiple pharmacologic and clinical endpoints Xueyuan Cao, Stanley Pounds, Susana Raimondi, James Downing, Raul Ribeiro, Jeffery Rubnitz, Jatinder Lamba P2 Integration of microRNA-mRNA interaction networks with gene expression data to increase experimental power Bernie J Daigle, Jr. P3 Designing and writing software for in silico subtractive hybridization of large eukaryotic genomes Deborah Burgess, Stephanie Gehrlich, John C Carmen P4 Tracking the molecular evolution of Pax gene Nicholas Johnson; Chandrakanth Emani P5 Identifying genetic differences in thermally dimorphic and state specific fungi using in silico genomic comparison Stephanie Gehrlich, Deborah Burgess, John C Carmen P6 Identification of conserved genomic regions and variation therein amongst Cetartiodactyla species using next generation sequencing Kalpani De Silva, Michael P Heaton, Theodore S Kalbfleisch P7 Mining physiological data to identify patients with similar medical events and phenotypes Teeradache Viangteeravat, Rahul Mudunuri, Oluwaseun Ajayi, Fatih Şen, Eunice Y Huang P8 Smart brief for home health monitoring Mohammad Mohebbi, Luaire Florian, Douglas J Jackson, John F Naber P9 Side-effect term matching for computational adverse drug reaction predictions AKM Sabbir, Sally R Ellingson P10 Enrichment vs robustness: A comparison of transcriptomic data clustering metrics Yuping Lu, Charles A Phillips, Michael A Langston P11 Deep neural networks for transcriptome-based cancer classification Rahul K Sevakula, Raghuveer Thirukovalluru, Nishchal K. Verma, Yan Cui P12 Motif discovery using K-means clustering Mohammed Sayed, Juw Won Park P13 Large scale discovery of active enhancers from nascent RNA sequencing Jing Wang, Qi Liu, Yu Shyr P14 Computationally characterizing genomic pipelines and benchmarking results using GATK best practices on the high performance computing cluster at the University of Kentucky Xiaofei Zhang, Sally R Ellingson P15 Development of approaches enabling the identification of abnormal gene expression from RNA-Seq in personalized oncology Naresh Prodduturi, Gavin R Oliver, Diane Grill, Jie Na, Jeanette Eckel-Passow, Eric W Klee P16 Processing RNA-Seq data of plants infected with coffee ringspot virus Michael M Goodin, Mark Farman, Harrison Inocencio, Chanyong Jang, Jerzy W Jaromczyk, Neil Moore, Kelly Sovacool P17 Comparative transcriptomics of three Acinetobacter baumanii clinical isolates with different antibiotic resistance patterns Leon Dent, Mike Izban, Sammed Mandape, Shruti Sakhare, Siddharth Pratap, Dana Marshall P18 Metagenomic assessment of possible microbial contamination in the equine reference genome assembly M Scotty DePriest, James N MacLeod, Theodore S Kalbfleisch P19 Molecular evolution of cancer driver genes Chandrakanth Emani, Hanady Adam, Ethan Blandford, Joel Campbell, Joshua Castlen, Brittany Dixon, Ginger Gilbert, Aaron Hall, Philip Kreisle, Jessica Lasher, Bethany Oakes, Allison Speer, Maximilian Valentine P20 Biorepository Laboratory Information Management System Naga Satya V Rao Nagisetty, Rony Jose, Teeradache Viangteeravat, Robert Rooney, David Hains