Searcharxiv⌕ Search

arXiv subjects

Afshin Sadeghi

Publications and source records attributed to Afshin Sadeghi.

7 recordsLinked to original sources

Do Probing and Steering Favor the Same Directions Within and Across Language Models?

A direction in a language model's hidden state can be used to probe whether a concept is present or to steer the model toward it. Although probe directions are often reused for steering, the direction that best detects a concept need not be the one that best changes it. We study this distinction in three instruction-tuned models from different families. We construct a one-parameter family of directions ranging from a covariance-adjusted discriminant to the mean difference, and evaluate probing and additive steering, which adds the direction to the hidden state, along the family at a fixed side effect. For truthfulness and sentiment, in every model, the best probe is not the best steering direction: steering consistently favors directions closer to the mean difference. This separation holds across 59 of 60 layer-model-concept combinations in a layer sweep, with matched probe and steering texts in all six model-concept combinations, and under a rotation-based steering operator. The exact steering optimum is less stable: for sentiment, an intermediate direction produces a 27-56% stronger steering effect than the mean difference, but this gain shrinks or disappears with matched texts and depends on the layer and the operator. The two roles also transfer differently across models. A probe transferred by a linear map fitted on neutral text retains 81-91% of the target model's above-chance probing performance, while the transferred mean difference retains less of its steering effect in 11 of 12 model pairs. This is a consistent tendency, although our predefined criteria classify only three pairs, and nine remain inconclusive. Finally, as an application on PopQA, steering toward "unknown" increases abstention by 18-44 percentage points on uncertain errors, versus 2-5 on confident errors, while rarely correcting errors in cases where a correct answer is reachable.

cs.CL↗

MDE: Multiple Distance Embeddings for Link Prediction in Knowledge Graphs

Over the past decade, knowledge graphs became popular for capturing structured domain knowledge. Relational learning models enable the prediction of missing links inside knowledge graphs. More specifically, latent distance approaches model the relationships among entities via a distance between latent representations. Translating embedding models (e.g., TransE) are among the most popular latent distance approaches which use one distance function to learn multiple relation patterns. However, they are mostly inefficient in capturing symmetric relations since the representation vector norm for all the symmetric relations becomes equal to zero. They also lose information when learning relations with reflexive patterns since they become symmetric and transitive. We propose the Multiple Distance Embedding model (MDE) that addresses these limitations and a framework to collaboratively combine variant latent distance-based terms. Our solution is based on two principles: 1) we use a limit-based loss instead of a margin ranking loss and, 2) by learning independent embedding vectors for each of the terms we can collectively train and predict using contradicting distance terms. We further demonstrate that MDE allows modeling relations with (anti)symmetry, inversion, and composition patterns. We propose MDE as a neural network model that allows us to map non-linear relations between the embedding vectors and the expected output of the score function. Our empirical results show that MDE performs competitively to state-of-the-art embedding models on several benchmark datasets.

cs.AI↗

Linking Physicians to Medical Research Results via Knowledge Graph Embeddings and Twitter

Informing professionals about the latest research results in their field is a particularly important task in the field of health care, since any development in this field directly improves the health status of the patients. Meanwhile, social media is an infrastructure that allows public instant sharing of information, thus it has recently become popular in medical applications. In this study, we apply Multi Distance Knowledge Graph Embeddings (MDE) to link physicians and surgeons to the latest medical breakthroughs that are shared as the research results on Twitter. Our study shows that using this method physicians can be informed about the new findings in their field given that they have an account dedicated to their profession.

cs.SI↗

Automatically Annotating Articles Towards Opening and Reusing Transparent Peer Reviews

An increasing number of scientific publications are created in open and transparent peer review models: a submission is published first, and then reviewers are invited, or a submission is reviewed in a closed environment but then these reviews are published with the final article, or combinations of these. Reasons for open peer review include giving better credit to reviewers and enabling readers to better appraise the quality of a publication. In most cases, the full, unstructured text of an open review is published next to the full, unstructured text of the article reviewed. This approach prevents human readers from getting a quick impression of the quality of parts of an article, and it does not easily support secondary exploitation, e.g., for scientometrics on reviews. While document formats have been proposed for publishing structured articles including reviews, integrated tool support for entire open peer review workflows resulting in such documents is still scarce. We present AR-Annotator, the Automatic Article and Review Annotator which employs a semantic information model of an article and its reviews, using semantic markup and unique identifiers for all entities of interest. The fine-grained article structure is not only exposed to authors and reviewers but also preserved in the published version. We publish articles and their reviews in a Linked Data representation and thus maximize their reusability by third-party applications. We demonstrate this reusability by running quality-related queries against the structured representation of articles and their reviews.

cs.IR↗

A Survey of User Expectations and Tool Limitations in Collaborative Scientific Authoring and Reviewing

Collaborative scientific authoring is increasingly being supported by software tools. Traditionally, desktop-based authoring tools had the most advanced editing features, allowed for more formatting options, and included more import/export filters. Web-based tools have excelled in their collaboration support. Recently, developers on both sides have been trying to close this gap by extending desktop-based tools to better support collaboration and by making web-based tools richer in functionality. To verify to what extent these developments actually meet the needs of researchers, we gathered precise requirements towards better tool support for scientific authoring and reviewing workflows by interviewing 213 users and studying a corpus of 27 documents. We present the design of the survey and interpret its results. The conclusion is that WYSIWYG and offline desktop authoring tools continue to be more popular among academics than text-based and online editors.

cs.DL↗

Decentralized creation of academic documents using a Network Attached Storage (NAS) server

Scholarly document creation continues to face various obstacles. Scholarly text production requires more complex word processors than other forms of texts because of the complex structures of citations, formulas and figures. The need for peer review, often single-blind or double-blind, creates needs for document management that other texts do not require. Additionally, the need for collaborative editing, security and strict document access rules means that many existing word processors are imperfect solutions for academics. Nevertheless, most papers continue to be written using Microsoft Word (Sadeghi et al. 2017). We here analyze some of the problems with existing academic solutions and then present an argument why we believe that running an open source academic writing solution for academic purposes, such as Fidus Writer, on a Network Attached Storage (NAS) server could be a viable alternative.

cs.DL↗

Opening Scholarly Communication in Social Sciences by Connecting Collaborative Authoring to Peer Review

The objective of the OSCOSS research project on "Opening Scholarly Communication in the Social Sciences" is to build a coherent collaboration environment that facilitates scholarly communication workflows of social scientists in the roles of authors, reviewers, editors and readers. This paper presents the implementation of the core of this environment: the integration of the Fidus Writer academic word processor with the Open Journal Systems (OJS) submission and review management system.

cs.DL↗