Coactive Learning

Shivaswamy, Pannaga; Joachims, Thorsten

doi:10.1613/jair.4539

Cited by 48 publications

(86 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where k is the action space dimensionality and H u * H is the Hessian of the exponent in the denominator of (19) around u * H . We obtain the optimal action u * H by solving the constrained optimization problem (see Appendix A-B):…”

Section: B Approximationmentioning

confidence: 99%

See 1 more Smart Citation

Quantifying Hypothesis Space Misspecification in Learning From Human–Robot Demonstrations and Physical Corrections

et al. 2020

View full text Add to dashboard Cite

Section: B Approximationmentioning

confidence: 99%

“…Since ∇C Φ D (ū) has a global minimum at u * H then ∇C Φ D (u * H ) = 0 and the denominator of Equation 19 can be rewritten as:…”

Section: Laplace Approximation In Equation (19)mentioning

confidence: 99%

Quantifying Hypothesis Space Misspecification in Learning From Human–Robot Demonstrations and Physical Corrections

et al. 2020

View full text Add to dashboard Cite

“…Larger sets of alternatives can be more informative, at the cost of a potentially higher cognitive cost for the user, as detailed in Section 4. • Coactive feedback: this was recently proposed by Shivaswamy and Joachims (2015) as an alternative to comparative feedback. It is a type of manipulative feedback, where the system provides a single recommendation and the user is asked to (slightly) improve it to better match her preferences.…”

Section: Preference Elicitationmentioning

confidence: 99%

“…The critiques are only requested at specific iterations, to minimize the user effort. We include a theoretical analysis of the CPP algorithm, derived from Shivaswamy and Joachims (2015), that elucidates the convergence properties of the algorithm.…”

Section: Introductionmentioning

confidence: 99%

Constructive Preference Elicitation

2018

View full text Add to dashboard Cite

When faced with large or complex decision problems, human decision makers (DM) can make costly mistakes, due to inherent limitations of their memory, attention, and knowledge. Preference elicitation tools assist the decision maker in overcoming these limitations. They do so by interactively learning the DM's preferences through appropriately chosen queries and suggesting high-quality outcomes based on the preference estimates. Most state-of-the-art techniques, however, fail in constructive settings, where the goal is to synthesize a custom or entirely novel configuration rather than choosing the best option among a given set of candidates. Many widespread problems are constructive in nature: customizing composite goods such as cars and computers, bundling products, recommending touristic travel plans, designing apartments, buildings, or urban layouts, etc. In these settings, the full set of outcomes is humongous and can not be explicitly enumerated, and the solution must be synthesized via constrained optimization. In this article, we describe recent approaches especially designed for constructive problems, outlining the underlying ideas and their differences as well as their limitations. In presenting them, we especially focus on novel issues that the constructive setting brings forth, such as how to deal with sparsity of the DM's preferences, how to properly frame the interaction, and how to achieve efficient synthesis of custom instances.

show abstract

“…Her only way of influencing the results is via the provided data or labels. For that reason Shivaswamy and Joachims [94] extended this towards coactive learning, where the teacher can also correct the learner during learning if necessary, providing a slightly improved but not necessarily optimal example as feedback.…”

Section: Interactive Data Analyticsmentioning

confidence: 99%

Interactive Data Analytics for the Humanities

Gurevych

Meyer

Binnig

et al. 2018

Computational Linguistics and Intelligent Text Processing

View full text Add to dashboard Cite

In this vision paper, we argue that current solutions to data analytics are not suitable for complex tasks from the humanities, as they are agnostic of the user and focused on static, predefined tasks with large-scale benchmarks. Instead, we believe that the human must be put into the loop to address small data scenarios that require expert domain knowledge and fluid, incrementally defined tasks, which are common for many humanities use cases. Besides the main challenges, we discuss existing and urgently required solutions to interactive data acquisition, model development, model interpretation, and system support for interactive data analytics. In the envisioned interactive systems, human users not only provide annotations to a machine learner, but train a model by using the system and demonstrating the task. The learning system will actively query the user for feedback, refine its model in real-time, and is able to explain its decisions. Our vision links natural language processing research with recent advances in machine learning, computer vision, and data management systems, as realizing this vision relies on combining expertise from all of these scientific fields. 1 Challenges in Analyzing Humanities Data Automated data analytics, aka. data mining and machine learning, is a key technology for enriching and interpreting data, making informed decisions, and developing new data-driven scientific methods across many disciplines in industry and academia. Although the potential of interactive problem solving was recognized early on [16], this field has not progressed very far beyond the initial work. In particular, interactive machine learning and data analytics have only recently received increased attention [98]. Current data analytics solutions focus predominantly on well-defined tasks that can be solved by processing large, homogeneous datasets available in a structured form. Consider for example recommender systems, which suggest new products based on the product's properties, the products that the customer has previously bought, and the collective behavior of the customer database [43]. The state-of-the-art relies on huge amounts of data-over one billion pairs of users and news items passively gathered-to train a deep neural network [28]. This may explain, why data analytics is conceived in a rather impersonal way, with algorithms working autonomously on passively collected data, although practice is quite the opposite. Most of the influence practitioners have, comes through interacting with data, including crafting the data and examining results. In the late 1990s, digitized data became widely available in the humanities as well. Since then, there has been a clear demand for data analytics approaches to tap into these

show abstract

Coactive Learning

Cited by 48 publications

References 28 publications

Quantifying Hypothesis Space Misspecification in Learning From Human–Robot Demonstrations and Physical Corrections

Quantifying Hypothesis Space Misspecification in Learning From Human–Robot Demonstrations and Physical Corrections

Constructive Preference Elicitation

Interactive Data Analytics for the Humanities

Contact Info

Product

Resources

About