A5001517374- Data Visualization and Analytics
- Topic Modeling
- Software Engineering Research
- Information and Cyber Security
- Big Data and Business Intelligence
- Open Source Software Innovations
- Team Dynamics and Performance
- Data Quality and Management
- Natural Language Processing Techniques
- Advanced Text Analysis Techniques
- Mobile Crowdsensing and Crowdsourcing
- Human-Automation Interaction and Safety
- Advanced Database Systems and Queries
- Software Engineering Techniques and Practices
- Multi-Agent Systems and Negotiation
- Ethics in Clinical Research
- Clinical practice guidelines implementation
- Domain Adaptation and Few-Shot Learning
- Electronic Health Records Systems
- Explainable Artificial Intelligence (XAI)
- Virtual Reality Applications and Impacts
- Multimodal Machine Learning Applications
- Species Distribution and Climate Change
- Machine Learning in Materials Science
- Military Strategy and Technology
Google (United States)
2023
Cerner (United States)
2023
Aptima (United States)
2014-2020
This work introduces Gemma, a family of lightweight, state-of-the art open models built from the research and technology used to create Gemini models. Gemma demonstrate strong performance across academic benchmarks for language understanding, reasoning, safety. We release two sizes (2 billion 7 parameters), provide both pretrained fine-tuned checkpoints. outperforms similarly sized on 11 out 18 text-based tasks, we present comprehensive evaluations safety responsibility aspects models,...
The goal of this article is to describe an integrated parallel process for the co-development written and computable clinical practice guidelines (CPGs) accelerate adoption increase impact guideline recommendations in practice. From February 2018 through December 2021, interdisciplinary work groups were formed after initial Kaizen event using expert consensus available literature, produced a 12-phase (IP). IP includes activities, resources, iterative feedback loops developing, implementing,...
Models are interpretable when machine learning (ML) practitioners can readily understand the reasoning behind their predictions. Ironically, little is known about ML practitioners' experience of discovering and adopting novel interpretability techniques in production settings. In a qualitative study with 18 at large technology company working text data, we found that despite varied tasks, experienced nearly identical challenges related to methods model analysis workflows. These stem from...
In recent years, advances in artificial intelligence (AI) have far outpaced our ability to understand and leverage them. no domain has this been more true than conversational agents (CAs). Transformer-based generative language models, such as GPT-2, significantly advance CAs' generate creative relevant content. It is critical start exploring collaboration with these CAs. paper, we focus on an initial step by enabling a human-augmented, AI-driven CA contribute panel discussion. Key questions...
We present Sequence Salience, a visual tool for interactive prompt debugging with input salience methods. Salience builds on widely used methods text classification and single-token prediction, extends this to system tailored complex LLM prompts. Our is well-suited long texts, expands previous work by 1) providing controllable aggregation of token-level the word, sentence, or paragraph level, making over inputs tractable; 2) supporting rapid iteration where practitioners can act results,...
We present ShieldGemma, a comprehensive suite of LLM-based safety content moderation models built upon Gemma2. These provide robust, state-of-the-art predictions risks across key harm types (sexually explicit, dangerous content, harassment, hate speech) in both user input and LLM-generated output. By evaluating on public internal benchmarks, we demonstrate superior performance compared to existing models, such as Llama Guard (+10.8\% AU-PRC benchmarks) WildCard (+4.3\%). Additionally, novel...
Recent advances in artificial intelligence have demonstrated that the future of work will be defined by collaborative human-machine teams. In order to effective, teams rely on context-aware systems enable collaboration. this paper, we present three lessons learned from past five years developing believe improve system design. First, semantic activity must captured, modeled, and analyzed reasoning across missions, actors, content. Second, require multiple, federated data stores optimize team...
Military planners and decision-makers face a number of challenges with the shift towards operating within diverse, multi-dimensional, unconventional environments. Leaders require deeper understanding broader social civil context in which operations occur, including underlying factors that contribute to instability drivers conflict. This is often derived through analysis textual data. Both traditional non-traditional sources – such as news articles, blog entries, tweets represent vast amount...
Spatial navigation of virtual environments often requires different categories actions. In this paper, we describe a mapping between forms and multimodal interactions to accomplish the We select define three navigation, identify characteristics those forms, that were selected match characteristics. developed these mappings within multidisplay, semi-immersive environment assembled from commodity components, with goal supporting use similar semiimmersive in operational settings. designing...
Recently, cyber reasoning systems demonstrated near-human performance characteristics when they autonomously identified, proved, and mitigated vulnerabilities in software during a competitive event. New research seeks to augment human vulnerability teams with system teammates collaborative work environments. However, the literature lacks concrete understanding of workflows practices, limiting designers’, engineers’, researchers’ ability successfully integrate these artificially intelligent...
We present a prototype visual analytics tool for facilitating analysis in collaborative team setting by providing flexible platform visualizing and exploring multiple linked dataseis as unified graph model. Our approach uses Layered Graph Model to connect afford querying across data. used this explore form hypotheses about the disappearance on Kronos Mini-Challenge 1, revealing some initial strengths weaknesses of approach.
Large language models (LLMs) can synthesize code from natural descriptions or by completing in-context. In this paper, we consider the ability of LLMs to code, at inference time, for a novel API not in its training data, and specifically examine impact different designs on ability. We find that: 1) examples model data seem facilitate use time; 2) hallucination is most common failure mode; 3) both prompt affect performance. light these findings, introduce concept Synthetic API: an designed be...