Alignment for Advanced Machine Learning Systems

This chapter surveys eight research areas organized around one question: As learning systems become increasingly intelligent and autonomous, what design principles can best ensure that their behavior is aligned with the interests of the operators? The chapter focuses on two major technical obstacles to AI alignment: the challenge of specifying the right kind of objective functions and the challenge of designing AI systems that avoid unintended consequences and undesirable behavior even in cases where the objective function does not line up perfectly with the intentions of the designers. The questions surveyed include the following: How can we train reinforcement learners to take actions that are more amenable to meaningful assessment by intelligent overseers? What kinds of objective functions incentivize a system to “not have an overly large impact” or “not have many side effects”? The chapter discusses these questions, related work, and potential directions for future research, with the goal of highlighting relevant research topics in machine learning that appear tractable today.

Download Full-text

Automated Reasoning for Explainable Artificial Intelligence

10.29007/4b7h ◽

2018 ◽

Author(s):

Maria Paola Bonacina

Keyword(s):

Artificial Intelligence ◽

Machine Learning ◽

Automated Reasoning ◽

Position Statement ◽

Short Paper ◽

Future Research ◽

Grand Challenge ◽

New Paradigm ◽

Research Areas ◽

Explainable Artificial Intelligence

Reasoning and learning have been considered fundamental features of intelligence ever since the dawn of the field of artificial intelligence, leading to the development of the research areas of automated reasoning and machine learning. This short paper is a non-technical position statement that aims at prompting a discussion of the relationship between automated reasoning and machine learning, and more generally between automated reasoning and artificial intelligence. We suggest that the emergence of the new paradigm of XAI, that stands for eXplainable Artificial Intelligence, is an opportunity for rethinking these relationships, and that XAI may offer a grand challenge for future research on automated reasoning.

Download Full-text

AI and the Future of Cyber Competition

10.51593/2020ca007 ◽

2021 ◽

Author(s):

Wyatt Hoffman

Keyword(s):

Machine Learning ◽

Real World ◽

Learning Systems ◽

Cyber Defense ◽

The Future ◽

To Come ◽

The Right

As states turn to AI to gain an edge in cyber competition, it will change the cat-and-mouse game between cyber attackers and defenders. Embracing machine learning systems for cyber defense could drive more aggressive and destabilizing engagements between states. Wyatt Hoffman writes that cyber competition already has the ingredients needed for escalation to real-world violence, even if these ingredients have yet to come together in the right conditions.

Download Full-text

Review of Data Mining Techniques and Parameters for Recommendation of Effective Adaptive E-Learning System

Collaborative Filtering Using Data Mining and Analysis - Advances in Data Mining and Database Management ◽

10.4018/978-1-5225-0489-4.ch001 ◽

2017 ◽

pp. 1-23 ◽

Cited By ~ 1

Author(s):

Renuka Mahajan

Keyword(s):

Data Mining ◽

Recommender System ◽

Learning Systems ◽

Learning System ◽

Future Research ◽

Comprehensive List ◽

Research Areas ◽

New Models ◽

E Learning ◽

Research Intensity

This chapter revolves around the synthesis of three research areas- data mining, personalization, recommendation systems and adaptive e-Learning systems. It also introduces a comprehensive list of parameters, extricated by reviewing the existing research intensity during the period of 2000 to October 2014, for understanding what should be essential parameters for adapting an e-learning. In general, we can consider and answer few questions to answer this body of literature ‘what' can be adapted? What can we adapt to? How do we adapt? This review tries to answer on ‘what' can be adapted. Thus, it advances earlier personalization studies. The gaps in the previous studies in building adaptive e-learning systems were also reviewed. It can help in designing new models for adaptation and formulating novel recommender system techniques. This will provide a foundation to industry experts and scientists for future research in adaptive e-learning.

Download Full-text

INCREMENTAL LEARNING IN BIOLOGICAL AND MACHINE LEARNING SYSTEMS

International Journal of Neural Systems ◽

10.1142/s0129065702001308 ◽

2002 ◽

Vol 12 (06) ◽

pp. 447-465 ◽

Cited By ~ 13

Author(s):

STEPHAN K. CHALUP

Keyword(s):

Machine Learning ◽

Neural Networks ◽

Artificial Neural Networks ◽

Side Effects ◽

Incremental Learning ◽

Learning Systems ◽

Experimental Results ◽

Recurrent Artificial Neural Networks ◽

Standard Learning

Incremental learning concepts are reviewed in machine learning and neurobiology. They are identified in evolution, neurodevelopment and learning. A timeline of qualitative axon, neuron and synapse development summarizes the review on neurodevelopment. A discussion of experimental results on data incremental learning with recurrent artificial neural networks reveals that incremental learning often seems to be more efficient or powerful than standard learning but can produce unexpected side effects. A characterization of incremental learning is proposed which takes the elaborated biological and machine learning concepts into account.

Download Full-text

Rationale Discovery and Explainable AI

10.3233/faia210341 ◽

2021 ◽

Author(s):

Cor Steging ◽

Silja Renooij ◽

Bart Verheij

Keyword(s):

Machine Learning ◽

Feature Detection ◽

State Of The Art ◽

High Accuracy ◽

Learning Systems ◽

Training Data ◽

Relevant Feature ◽

Explainable Ai ◽

The Right ◽

The Impact

The justification of an algorithm’s outcomes is important in many domains, and in particular in the law. However, previous research has shown that machine learning systems can make the right decisions for the wrong reasons: despite high accuracies, not all of the conditions that define the domain of the training data are learned. In this study, we investigate what the system does learn, using state-of-the-art explainable AI techniques. With the use of SHAP and LIME, we are able to show which features impact the decision making process and how the impact changes with different distributions of the training data. However, our results also show that even high accuracy and good relevant feature detection are no guarantee for a sound rationale. Hence these state-of-the-art explainable AI techniques cannot be used to fully expose unsound rationales, further advocating the need for a separate method for rationale evaluation.

Download Full-text

Artificial Intelligence and Machine Learning: An Instructor’s Exoskeleton in the Future of Education

Innovative Learning Environments in STEM Higher Education - SpringerBriefs in Statistics ◽

10.1007/978-3-030-58948-6_5 ◽

2021 ◽

pp. 79-105

Author(s):

Stephanie E. August ◽

Audrey Tsaima

Keyword(s):

Artificial Intelligence ◽

Machine Learning ◽

Learning Systems ◽

Future Research ◽

Technology Support ◽

Student Knowledge ◽

Customized Learning ◽

Future Of Education ◽

Authentic Learning Experiences ◽

The Impact

AbstractThe role of artificial intelligence in US education is expanding. As education moves toward providing customized learning paths, the use of artificial intelligence (AI) and machine learning (ML) algorithms in learning systems increases. This can be viewed as growing metaphorical exoskeletons for instructors, enabling them to provide a higher level of guidance, feedback, and autonomy to learners. In turn, the instructor gains time to sense student needs and support authentic learning experiences that go beyond what AI and ML can provide. Applications of AI-based education technology support learning through automated tutoring, personalizing learning, assessing student knowledge, and automating tasks normally performed by the instructor. This technology raises questions about how it is best used, what data provides evidence of the impact of AI and ML on learning, and future directions in interactive learning systems. Exploration of the use of AI and ML for both co-curricular and independent learnings in content presentation and instruction; interactions, communications, and discussions; learner activities; assessment and evaluation; and co-curricular opportunities provide guidance for future research.

Download Full-text

Fairness in Algorithmic Decision-Making: Applications in Multi-Winner Voting, Machine Learning, and Recommender Systems

Algorithms ◽

10.3390/a12090199 ◽

2019 ◽

Vol 12 (9) ◽

pp. 199 ◽

Cited By ~ 3

Author(s):

Yash Raj Shrestha ◽

Yongjie Yang

Keyword(s):

Machine Learning ◽

Decision Making ◽

Objective Function ◽

Recommender Systems ◽

Minority Groups ◽

Ethical Issues ◽

Fixed Number ◽

Future Research ◽

Design Of Algorithms ◽

Voting Machine

Algorithmic decision-making has become ubiquitous in our societal and economic lives. With more and more decisions being delegated to algorithms, we have also encountered increasing evidence of ethical issues with respect to biases and lack of fairness pertaining to algorithmic decision-making outcomes. Such outcomes may lead to detrimental consequences to minority groups in terms of gender, ethnicity, and race. As a response, recent research has shifted from design of algorithms that merely pursue purely optimal outcomes with respect to a fixed objective function into ones that also ensure additional fairness properties. In this study, we aim to provide a broad and accessible overview of the recent research endeavor aimed at introducing fairness into algorithms used in automated decision-making in three principle domains, namely, multi-winner voting, machine learning, and recommender systems. Even though these domains have developed separately from each other, they share commonality with respect to decision-making as an application, which requires evaluation of a given set of alternatives that needs to be ranked with respect to a clearly defined objective function. More specifically, these relate to tasks such as (1) collectively selecting a fixed number of winner (or potentially high valued) alternatives from a given initial set of alternatives; (2) clustering a given set of alternatives into disjoint groups based on various similarity measures; or (3) finding a consensus ranking of entire or a subset of given alternatives. To this end, we illustrate a multitude of fairness properties studied in these three streams of literature, discuss their commonalities and interrelationships, synthesize what we know so far, and provide a useful perspective for future research.

Download Full-text

Can machine learning extract the mechanisms controlling phytoplankton growth from large-scale observations? – A proof-of-concept study

Biogeosciences ◽

10.5194/bg-18-1941-2021 ◽

2021 ◽

Vol 18 (6) ◽

pp. 1941-1970

Author(s):

Christopher Holder ◽

Anand Gnanadesikan

Keyword(s):

Machine Learning ◽

Large Scale ◽

Phytoplankton Growth ◽

Qualitative Assessment ◽

Future Research ◽

Proof Of Concept ◽

Physiological Mechanisms ◽

Concept Study ◽

Phytoplankton Growth Rate ◽

The Right

Abstract. A key challenge for biological oceanography is relating the physiological mechanisms controlling phytoplankton growth to the spatial distribution of those phytoplankton. Physiological mechanisms are often isolated by varying one driver of growth, such as nutrient or light, in a controlled laboratory setting producing what we call “intrinsic relationships”. We contrast these with the “apparent relationships” which emerge in the environment in climatological data. Although previous studies have found machine learning (ML) can find apparent relationships, there has yet to be a systematic study examining when and why these apparent relationships diverge from the underlying intrinsic relationships found in the lab and how and why this may depend on the method applied. Here we conduct a proof-of-concept study with three scenarios in which biomass is by construction a function of time-averaged phytoplankton growth rate. In the first scenario, the inputs and outputs of the intrinsic and apparent relationships vary over the same monthly timescales. In the second, the intrinsic relationships relate averages of drivers that vary on hourly timescales to biomass, but the apparent relationships are sought between monthly averages of these inputs and monthly-averaged output. In the third scenario we apply ML to the output of an actual Earth system model (ESM). Our results demonstrated that when intrinsic and apparent relationships operate on the same spatial and temporal timescale, neural network ensembles (NNEs) were able to extract the intrinsic relationships when only provided information about the apparent relationships, while colimitation and its inability to extrapolate resulted in random forests (RFs) diverging from the true response. When intrinsic and apparent relationships operated on different timescales (as little separation as hourly versus daily), NNEs fed with apparent relationships in time-averaged data produced responses with the right shape but underestimated the biomass. This was because when the intrinsic relationship was nonlinear, the response to a time-averaged input differed systematically from the time-averaged response. Although the limitations found by NNEs were overestimated, they were able to produce more realistic shapes of the actual relationships compared to multiple linear regression. Additionally, NNEs were able to model the interactions between predictors and their effects on biomass, allowing for a qualitative assessment of the colimitation patterns and the nutrient causing the most limitation. Future research may be able to use this type of analysis for observational datasets and other ESMs to identify apparent relationships between biogeochemical variables (rather than spatiotemporal distributions only) and identify interactions and colimitations without having to perform (or at least performing fewer) growth experiments in a lab. From our study, it appears that ML can extract useful information from ESM output and could likely do so for observational datasets as well.

Download Full-text

Intelligent COVID-19 Forecasting, Diagnoses and Monitoring Systems: A Survey

10.36227/techrxiv.15172488 ◽

2021 ◽

Author(s):

Nasreen Anjum ◽

Amna Asif, ◽

Mehreen Kiran ◽

Fouzia Jabeen ◽

Zhaohui Yang ◽

...

Keyword(s):

Machine Learning ◽

Learning Algorithms ◽

Herd Immunity ◽

Machine Learning Algorithms ◽

Future Research ◽

Monitoring Systems ◽

Extensive Literature ◽

Awareness Campaigns ◽

Research Areas ◽

Wide Range

<div>To date, the novel Corona virus (SARS-CoV-2) has infected millions and has caused the deaths of thousands of people around the world. At the moment, five antibodies, two from China, two from the U.S., and one from the UK, have already been widely utilized and numerous vaccines are under the trail process. In order to reach herd immunity, around 70% of the population would need to be inoculated. It may take several years to hinder the spread of SARS-CoV-2. Governments and concerned authorities have taken stringent measurements such as enforcing partial, complete, or smart lockdowns, building temporary medical facilities, advocating social distancing, and mandating masks in public as well as setting up awareness campaigns. Furthermore, there have been massive efforts in various research areas and a wide variety of tools, technologies and techniques have been explored and developed to combat the war against this pandemic. Interestingly, machine learning algorithms and internet of Things (IoTs) technology are the pioneers in this race. Up till now, several real-time and intelligent COVID-19 forecasting, diagnosing, and monitoring systems have been proposed to tackle the COVID-19 pandemic. In this article based on our extensive literature review, we provide a taxonomy based on the intelligent COVID-19 forecasting, diagnosing, and monitoring systems. We review the available literature extensively under the proposed taxonomy and have analyzed a significantly wide range of machine learning algorithms and IoTs which can be used in predicting the spread of COVID-19 and in diagnosing and monitoring the infected individuals. Furthermore, we identify the challenges and also provide our vision about the future research on COVID-19.</div>

Download Full-text

The Conditional Entropy Bottleneck

Entropy ◽

10.3390/e22090999 ◽

2020 ◽

Vol 22 (9) ◽

pp. 999 ◽

Cited By ~ 3

Author(s):

Ian Fischer

Keyword(s):

Machine Learning ◽

Objective Function ◽

Failure Modes ◽

Conditional Entropy ◽

Learning Systems ◽

Training Data ◽

Deterministic Models ◽

Information Bottleneck ◽

Adversarial Examples

Much of the field of Machine Learning exhibits a prominent set of failure modes, including vulnerability to adversarial examples, poor out-of-distribution (OoD) detection, miscalibration, and willingness to memorize random labelings of datasets. We characterize these as failures of robust generalization, which extends the traditional measure of generalization as accuracy or related metrics on a held-out set. We hypothesize that these failures to robustly generalize are due to the learning systems retaining too much information about the training data. To test this hypothesis, we propose the Minimum Necessary Information (MNI) criterion for evaluating the quality of a model. In order to train models that perform well with respect to the MNI criterion, we present a new objective function, the Conditional Entropy Bottleneck (CEB), which is closely related to the Information Bottleneck (IB). We experimentally test our hypothesis by comparing the performance of CEB models with deterministic models and Variational Information Bottleneck (VIB) models on a variety of different datasets and robustness challenges. We find strong empirical evidence supporting our hypothesis that MNI models improve on these problems of robust generalization.

Download Full-text