Session 1: Plenary Themes in Discovery Informatics Science Has a Never-ending Thirst for Technology Computing as a substrate for science in innovative ways Ongoing investments in cyberinfrastructure have a tremendous impact in scientific discoveries Shared high end instruments High performance computing Distributed services Data management Virtual organizations These investments are extremely valuable for science, but do not address many aspects of science Further Science Needs Emphasis has been on data and computation, not so much on models Need to support model formulation and testing is missing Models should be related to data (observed or simulated) Emphasize insight and understanding From correlations to causality and explanation Developing tools for the full discovery process and using tools for the discovery process Tools that help you do new things vs tools that help you do things better Further Science Needs Many aspects of the scientific process could be improved Some are not addressed by CI (eg literature search, reasoning about models) Others could benefit from new approaches (eg capturing metadata) Effort is significant Many scientists do not have the resources or inclination to benefit from CI How do you create a culture in which science stays timely in its use of CI? Discipline-specific services make it harder to cross bounds Methods and process for being able to work with scientists Further Science Needs Integration is important and far from being a solved problem Integration across science domains Integration within a domain Connecting tools and technologies to the practice of science Most science is done local, need to respond accordingly (e.g., how do you support your student, get tenure) How to reduce the impedance mismatch between cognition and practice The “long tail” of science – most of science is not big science nor big data CI can transform all elements of the discovery timeline Further Science Needs User-centered design Usability Functionality What are metrics for success Adoption by others? Characterization of domains and facets that impact discovery informatics is still not understood You can’t get this by asking the scientists What are equivalent classes of domains as they pertain to CI Need to treat domain scientists, social scientists, and computer scientists on equal footing Emerging Movement? A movement for scientist-centered system design? A movement to focus on the “human processor bottleneck”? Human cognitive capacity is flat (or at best getting slightly linearly), while other dimensions of computing have grown exponentially A movement for non-centralized science? (“long tail” of science (on multiple dimensions) aka “dark matter” of science; small science vs big; small data vs large) A movement to improve the use of mundane technology in science practice? A movement to lower the learning curve in infrastructure? There will be some curve, but it is smaller and the same no matter what you need to access eg web infrastructure is a good example What is Discovery Informatics We should come back to a definition later in the meeting Some possible defining characteristics: Small data science still has a major role to play Complements big data science Much of science is largely local Complements science at larger scales Big data science can be seen as a movement to more centralized science The “long tail” of scientists are still largely underserved The “long tail” of scientific questions still has rudimentary technology Spreadsheets are still in widespread use Many valuable datasets are never integrated to address aggregate questions Discovery is a social endeavor Socio-technical systems to support ad-hoc collaborations Enable routine unexpected or indirect interactions among scientists eg, unanticipated data sharing DI: Automating and enhancing scientific processes at all levels? DI: Empowering individual researchers through local infrastructure? Do Scientific Discoveries Result from Special Kinds of Scientific Activities? Perhaps, but we do not need to address this question if we can agree to consider discoveries in a continuum The more the scientific processes are improved, the more the discovery processes are improved The more we empower scientists to cope with more complex models (larger scope, broader coverage), the more the discovery processes are improved The more we open access of potential contributors to scientific processes, the more the discovery processes are improved Discovery Informatics: Why Now Discovery informatics as “multiplicative science”: Investments in this area will have multiplicative gains as they will impact all areas of science and engineering Multiplicative in the dimension of the “human bottleneck” Could address current redundancy in {bio|geo|eco|…}informatics Discovery informatics will empower the public: Society is ready to participate in scientific activities and discovery tools can capture scientific practices “Personal data” will give rise to “personal science” I study my genes, my medical condition, my backyard’s ecosystem Volunteer donations of funds and time are now commonplace Enable donations of more intellectual contributions and insights Discovery informatics will enable lifelong learning and training of future workforce in all areas of science Focuses on usable tools that encapsulate, automate, and disseminate important aspects of state-of-the-art scientific practice Discovery Informatics: Why Now Scope to include engineering, medicine Science too big to fit in your head all at one time Need computation to help understand it Current process of conducting science in all areas is utterly broken, often reinventing processes year after year Science are more willing to adopt and collaborate Three Major Themes in Discovery Informatics IN THIS SESSION: 3) Social computing for science 1) Improving the discovery process For each theme: 1. Why important to discuss 2. State of the art (where is it published) 3. Topics Focus is on coming up 2) Learning models from science data as a group with topics that each breakout should elaborate Bring up a topic not yet listed but do not dwell on it THEME 1: Improving the Experimentation and Discovery Process Unprecedented complexity of scientific enterprise Is science stymied by the human bottleneck? What aspects of the process could be improved, e.g.: Managing publications through natural language technologies Capturing current knowledge through ontologies and models Multi-step data analysis through computational workflows Process reproducibility and reuse through provenance Data collection and analysis through integrated robotics Data sharing through Semantic Web Cross-disciplinary research through collaborative interfaces Result understanding through visualization THEME 2: Learning Models from Science Data Complexity of models and complexity of data analysis Data analysis activities placed in a larger context Comprehensive treatment of data to models to hypotheses cycle Using models to drive data collection activities Preparing data in service of model formation and hypothesis testing Selecting relevant features for model development Highlighting interesting behaviors and unusual results THEME 3: Social Computing for Science Multiplicative gains through broadening participation Some challenges require it, others can significantly benefit Managing human contributions What scientific tasks could be handled How can tasks be organized to facilitate contributions Can reusable infrastructure be developed Can junior researchers, K-12 students, and the public take more active roles in scientific discoveries Three Major Themes 3) Social computing for discovery 1) Improving the discovery process 2) Learning science models from data Improving the Discovery Process: Why Characterizing what the discovery process is Current processes are in many ways inefficient / less effective Manual data analysis Reproducibility is too costly Literature is vast and unmanageable … Improving the Discovery Process: What is the State of the Art Workflow systems Automate many aspects of data analysis, make it reproducible/reusable Emerging provenance standards (OPM, W3C’s PROV) Augmenting scientific publications with workflows Creating knowledge bases from publications Ontological annotations of articles including claims and evidence Text mining to extract assertions to create knowledge bases Reasoning with knowledge bases to suggest or check hypotheses Visualization 3 separate fields: scientific visualization, information visualization, and visual analytics “design studies” Combining visualizations with other data Improving the Discovery Process: What is the State of the Art What is the state of the art of what’s currently used in science? Opening data and models Visualization not just of data, but also models and relationships between models Improving the Discovery Process: Discussion Topics (I) Automation of discovery processes What is possible and unlikely in near/longer term Representations are key to discovery, hard to engineer change of representation in a system Challenge is to find the right division of labor between human and computer User-centered design Automation should come with suitable explanations Of processes, models, data, etc. Designing tools for the individual scientist (the “long tail”) Improving the Discovery Process: Discussion Topics (II) Workflows Understand barriers to widespread practice Have they reached the tipping point of usability vs pain? Workflow reuse across labs, across workflow systems Are workflows useful? What can we learn from workflows in non-science domains? Text extraction / generation Annotating publications Improving the Discovery Process: Discussion Topics (III) Visualizations could help maximize the bandwidth of what humans can assimilate Visualization Do scientists know what they want? Scientists seem to prefer interaction, ie, control over the visualization, rather than automatic visualizations Active co-creation of visualization helps scientists Domain specification / requirements extraction Centrality of knowledge representations (means to an end) Data Processes Reuse, open access, dynamic Enabling integrated representation, reasoning, and learning Risk of not being pertinent to some areas of science From Models to Data and Back Again: Why Need to integrate better data with models and sense-making Semantic integration to enable reasoning Linking claims to experimental designs to data Interpreting data is a cognitive social process, aided by visualizations that integrate context into the data How do we integrate prior knowledge, formalisms scientists use, how do we update knowledge/formalisms Generating useful data is a bottleneck, generating lots of models is easy, should leverage this Need to help scientists to evaluate models Learning “Models” from Data: What is the State of the Art Cognitive science studies of discovery and insight The role of effective problem representations The challenges of programming representation change Computational discovery Model-based reasoning Causality Temporal dependency analysis Design of quasi-experiments Spatial and temporal data Variability, multi-scale, Sensor noise Quality control Sensor noise vs actual phenomena Learning Models from Data: Discussion Topics (I) Integrating better models/knowledge and data Model-guided data collection Collect data based on goals Observations guiding the revision of models Explaining findings and revising models and knowledge Visualizations that combine models and data Deriving stuff from data Enable causal connections across diverse data sources Causal relations co-existing with gaps and conflicts stands in the way to more unified databases Models / patterns / laws? Importance of uncertainty, quality, utility From models to use Connecting computer simulations and model building from data HPC, simulation, and modeling from data should be connected Learning Models from Data: Discussion Topics (II) Learning models that are communicable Potential for unifying models and associated tools for doing so ML has a lot of theoretical results that have not yet been made useful more broadly Need to be more usable/accessible Particularly in social sciences Not always easy to apply to big data Learning Models from Data: Discussion Topics (III) Incentivizing digital resource sharing to enable discoveries Privacy and security: data being misused or not appropriately credited The social sciences are a particularly promising area for discovery informatics, and what would facilitate this Digital resource curation as a social issue Verification (of models, conclusions, data, explanations, etc.) Social Computing: Why Many valuable datasets lack appropriate metadata Labels, data characteristics and properties, etc. Human computation has beaten best of breed algorithms Social agreement accelerates data sharing Public interest in participating in scientific activity Community assessment of models, knowledge, etc. Concretizing elements that were mushy in the past Mixed-initiative processes – humans exceed machine in many areas, so we need to assimilate them for the things that they do better Harness knowledge about what makes online communities (including, e.g., Wikipedia) work well or poorly Role of incentives, motivation, in bringing people together to do science Social Computing: What is the State of the Art Very different manifestations: Collecting data (eg pictures of birds) Labeling data (eg Galaxy Zoo) Computations (eg Foldit) Elaborate human processes (eg theorem proving) Bringing people and computing together in complementary ways Social Computing: Discussion Topics (I) Several names: is there a distinction Crowdsourcing, citizen science, Designing the system Roles: peers, senior researchers, automation Incentives Training Platforms and infrastructure (using clouds right, social web platforms) Incorporating semantic information and metadata Expertise finding New modalities for peer review, scholarly communication Social Computing: Discussion Topics (II) Defining workflows with more elaborate processes that mix human processing with computer processing Humans to do more complex tasks Can facilitate reproducibility Enticing people to participate while ensuring quality Some existing systems should be revisited to be designed as social systems Workflow libraries and reuse tools Data curation tools Open software Social Computing: Discussion Topics (III) Systems that enable collaborations that are not deliberate but ad-hoc Opportunistic partnerships Unexpected uses of data Systems that support a marketplace of ideas and track credit New ideas/discoveries are often seen as a threat to the status quo, how do we facilitate integration Empower people to share ideas on a problem while credited Incentive structures for new models of scholarly communication, such as blogs