Spain

&quot;The goal of the work presented in this paper is to provide techniques to challenge the results of an algorithmic decision system relying on machine learning. We highlight the differences between explanations and justifications and outline a framework to generate evidence to support or to dismiss challenges. We also present the results of a preliminary study to assess users’ perception of the different types of challenges proposed here and their benefits to detect incorrect results.
&quot;

ECAI 2020

Towards a framework for challenging ML-based decisions

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers. 
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities. This includes talks to attend in person, talks to attend remotely, breaks with colleagues and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


AAAI 2023

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

"The goal of the work presented in this paper is to provide techniques to challenge the results of an algorithmic decision system relying on machine learning. We highlight the differences between explanations and justifications and outline a framework to generate evidence to support or to dismiss challenges. We also present the results of a preliminary study to assess users’ perception of the different types of challenges proposed here and their benefits to detect incorrect results.
"

technical paper

ECAI2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI, including:




**An extensive programme of workshops, tutorials, and invited speakers**

**The Prestigious Applications of Intelligent Systems conference (PAIS)**

**The Frontiers in AI track sessions**

**The Lab To Market event**

**The Women in AI time**

**The EU Challenges forum**

… and many more!

ECAI 2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI.

Previous work in the social sciences, psychology and linguistics has show that liars have some control over the content of their stories, however their underlying state of mind may "leak out" through the way that they tell them. To the best of our knowledge, no previous systematic effort exists in order to describe and model deception language for Brazilian Portuguese. To fill this important gap, we carry out a pioneering empirical linguistic study on false statements in Brazilian news. We methodically analyze linguistic features using the Fake.Br corpus, which includes both fake and true news. The results show that they present substantial lexical, syntactic and semantic variations, as well as punctuation and emotion distinctions.

Studying Dishonest Intentions in Brazilian Portuguese Texts

"This research presents an analysis of AI adoption based on the analysis of the ProgrammableWeb directory. ProgrammableWeb.com is known as the Web’s most popular open Application programming interfaces (APIs) repository. Ever since it was founded in 2005, it has been chronicling the daily evolution of the global API, and it is the most relied-on directory when it comes to discovering and searching for APIs. With owner’s permission, we downloaded the ProgrammableWeb directory dataset, containing 21609 APIs, 1641 libraries, 15329 SDKs, 7629 mashups, and 10989 sample source codes. These resources were analysed to have an indication/measure on the AI adoption related to the API domain. Overall, the analysis showed increasing artificial intelligence adoption
trends from 2005 onwards."

Tracking of Artificial Intelligence Adoption in ProgrammableWeb Directory

"In this paper we introduce a methodology for identifying early warning signs of transformative progress in AI, to aid anticipatory governance and research prioritisation. We propose using expert elicitation methods to identify milestones in
AI progress, followed by collaborative causal mapping to identify key milestones which underpin several others. We call these key milestones ‘canaries’ based on the colloquial phrase ‘canary in a coal mine’ to describe advance warning of an extreme event: in this case, advance warning of transformative AI. After describing and motivating our proposed methodology, we present results from an initial implementation to identify canaries for progress towards high-level machine intelligence (HLMI). We conclude by discussing the limitations of this method, possible future improvements, and how we hope it can be used to improve monitoring of future risks from AI progress. "

Canaries in Technology Mines: Warning Signs of Transformative Progress in AI

"Competitions are a proven and cost-effective method to quickly develop new disruptive technologies for new markets. However, artificial intelligence and robotics competitions are usually organized independently, in silos. To get the most out of competitions, evaluation of robots should be modular (tasks are designed to assess independently different technological building blocks of the robotic systems) and open (evaluation tools and data must be publicly and freely available online). This not only meets the benchmarking needs of end users (mostly interested in performance measurements of the complete system), but also those of integrators and developers of intelligent components (camera, lidar, radar, actuators, etc.).
With this objective in mind, the H2020 METRICS project (2020-2023) organizes competitions in four application areas (Healthcare, Infrastructure Inspection and Maintenance, Agri-Food, and Agile Production) relying on both physical testing facilities (field evaluation campaign) and virtual testing facilities (data-based evaluation campaign) to mobilize, in addition to the European robotics community, the artificial intelligence one.
This presentation outlines this approach and paves the way for a new robotics and artificial intelligence competition paradigm."

AI evaluation campaigns during robotics competitions: the METRICS paradigm

"The ROSE Challenge is the first robotics and artificial intelligence competition worldwide to implement a third-party performance evaluation of intra-row weeding robots in real conditions, under comparable conditions, to guarantee a credible and objective assessment of their effectiveness. This article reports on the design and
validation of testing facilities for this competition, which presents a particular complexity : the experiments take place outdoors and act on living things (crops and weeds). Moreover, it implies to guarantee repeatable experimental conditions for a comparable and equitable evaluation. The article also discusses the opportunity that these competitions represent to define testing facilities in a consensual way.
The method it proposes is very widely applicable to different fields of intelligent systems applications."

Design and validation of testing facilities for weeding robots as part of ROSE challenge

"Recent progress in machine learning, on predictive tasks in particular, is usually claimed on the basis of performance comparisons.
Due to the limitation of application scenarios and computational resources, most people cannot evaluate their work on a large variety of tasks and datasets, and it has become tricky for both reviewers and readers to verify individual performance gains for a given approach.
In this paper we investigate possible approaches to achieve better efficiency on model benchmarking. 
For a large collection of datasets, rather than training and testing a given approach on every individual dataset, we seek methods that allow us to pick only a few representative datasets to quantify the model's goodness, from which to extrapolate to performance on other datasets. 
To this end, we adopt existing approaches from psychometrics: specifically, Item Response Theory and Adaptive Testing. Both are well-founded frameworks designed for educational tests.
We propose certain modifications following the requirements of machine learning experiments, and present some initial results and insights in this paper."

Towards Efficient and Robust Model Benchmarks with Item Response Theory and Adaptive Testing

We consider the issue of AI (Artificial Intelligence) evaluation in the light of risk management for the proper marketing of machinery. When a physical device is driven by AI, the question arises as to how to manage the risks associated with the autonomous functions. Indeed, the risk of physical damage to persons or property in the event of a ``wrong'' decision must be properly assessed. The study presents the advances needed to carry out relevant risk analyses for the deployment of safe intelligent machines, and offers good practice that leverages the knowledge in AI evaluation to optimize the safety of these machines.

Risk assessment of artificial intelligence in autonomous machines

This paper argues that at a certain point in research toward AGI, the problem may become well-enough theorized that a clear roadmap exists for achieving it, such that a Manhattan Project- like effort could greatly shorten the time to completion. If state actors perceive that this threshold has been crossed, their incentives around openness and international cooperation may shift rather suddenly, with serious implications for AI risks and the stability of international AI governance regimes. The paper characterizes how such a ‘runway’ period would be qualitatively different from preceding stages of AI research, and accordingly proposes a research program aimed at assessing how close the field of AI is to such a threshold—that is, it calls for the formulation of a ‘roadmap to the roadmap.’

Roadmap to a Roadmap: How Could We Tell When AGI is a ‘Manhattan Project’ Away?

The widespread use of experimental benchmarks in AIresearch  has  created  new  competition  and  collaboration  dynamics that are still poorly understood. In this paper we provide an innovative methodology to explore this dynamics and analyse the way different entrants in these competitions, from academia to tech giants,behave and react depending on their own or others’ achievements.We perform an analysis of over twenty popular benchmarks in AI,linking their underlying research papers. We identify links between researchers and institutions (i.e., communities) beyond the standard co-authorship relations, and we explore a series of hypotheses about their behaviour as well as some aggregated results in terms of activity, breakthroughs and efficiency. As a result, we detect and characterise the dynamics of research communities at different levels of abstraction, including organisation, affiliation, trajectories, results and activity.

The Scientometrics of AI Benchmarks: Unveiling the Underlying Mechanics of AI Research

The performance of the machine learning algorithms can be empirically evaluated via test campaigns: conventional testing approaches allow to obtain, from annotated databases, a first estimate of the operating environment of artificial intelligence systems. However, as those data are hardly I .i.d, additional scores could be useful to offer complementary performance assessment typically for intelligent systems performing critical functions. This paper proposes an approach to generate, without annotation phase, realistic corner cases to complete the exploration of the operating environment. The results of the proof of concept provided for optical character recognition highlight the interest of test data generation and automatic scenario selection methods, including GAN and adversarial approaches, to address this evaluation need.

Towards a framework for challenging ML-based decisions

Downloads

Next from ECAI 2020

Studying Dishonest Intentions in Brazilian Portuguese Texts

Similar lecture

Rethinking Rotation Invariance with Point Cloud Registration

Towards a framework for challenging ML-based decisions

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ECAI 2020

Studying Dishonest Intentions in Brazilian Portuguese Texts

Similar lecture

Rethinking Rotation Invariance with Point Cloud Registration

Downloads