Workshop: Algorithmic Fairness through the lens of Causality and Robustness Mon 13 Dec 01:00 a.m.
Trustworthy machine learning (ML) encompasses multiple fields of research, including (but not limited to) robustness, algorithmic fairness, interpretability and privacy. Recently, relationships between techniques and metrics used across different fields of trustworthy ML have emerged, leading to interesting work at the intersection of algorithmic fairness, robustness, and causality.
On one hand, causality has been proposed as a powerful tool to address the limitations of initial statistical definitions of fairness. However, questions have emerged regarding the applicability of such approaches in practice and the suitability of a causal framing for studies of bias and discrimination. On the other hand, the Robustness literature has surfaced promising approaches to improve fairness in ML models. For instance, parallels can be shown between individual fairness and local robustness guarantees. In addition, the interactions between fairness and robustness can help us understand how fairness guarantees hold under distribution shift or adversarial/poisoning attacks.
After a first edition of this workshop that focused on causality and interpretability, we will turn to the intersectionality between algorithmic fairness and recent techniques in causality and robustness. In this context, we will investigate how these different topics relate, but also how they can augment each other to provide better or more suited definitions and mitigation strategies for algorithmic fairness. We are particularly interested in addressing open questions in the field, such as:
- How can causally grounded fairness methods help develop more robust and fair algorithms in practice?
- What is an appropriate causal framing in studies of discrimination?
- How do approaches for adversarial/poisoning attacks target algorithmic fairness?
- How do fairness guarantees hold under distribution shift?
5th Workshop on Meta-Learning Mon 13 Dec 03:00 a.m.
Recent years have seen rapid progress in meta-learning methods, which transfer knowledge across tasks and domains to efficiently learn new tasks, optimize the learning process itself, and even generate new learning methods from scratch. Meta-learning can be seen as the logical conclusion of the arc that machine learning has undergone in the last decade, from learning classifiers, to learning representations, and finally to learning algorithms that themselves acquire representations, classifiers, and policies for acting in environments. In practice, meta-learning has been shown to yield new state-of-the-art automated machine learning methods, novel deep learning architectures, and substantially improved one-shot learning systems. Moreover, to improve one’s own learning capabilities through experience can also be viewed as a hallmark of intelligent beings, and neuroscience shows a strong connection between human and reward learning and the growing sub-field of meta-reinforcement learning.
Workshop: OPT 2021: Optimization for Machine Learning Mon 13 Dec 03:15 a.m.
OPT 2021 will bring experts in optimization to share their perspectives while leveraging crossover experts in ML to share their views and recent advances. OPT 2021 honors this tradition of bringing together people from optimization and from ML in order to promote and generate new interactions between the two communities.
To foster the spirit of innovation and collaboration, a goal of this workshop, OPT 2021 will focus the contributed talks on research in “Beyond Worst-case Complexity”. Classical optimization analyses measure the performances of algorithms based on (1). the computation cost and (2). convergence for any input into the algorithm. Yet algorithms with worse traditional complexity (e.g. SGD and its variants, ADAM, etc), are increasingly popular in practice for training deep neural networks and other ML tasks. This leads to questions such as what are good modeling assumptions for ML problems to measure an optimization algorithm’s success and how can we leverage these to better understand the performances of known (and new) algorithms. For instance, typical optimization problems in ML may be better conditioned than their worst-case counterparts in part because the problems are highly structured and/or high-dimensional (large number of features/samples). One could leverage this observation to design algorithms with better “average-case” complexity. Moreover, increasing research seems to indicate an intimate connection between the optimization algorithm and how well it performs on the test data (generalization). This new area of research in ML and its deep ties to optimization warrants a necessary discussion between the two communities. Specifically, we aim to continue the discussion on the precise meaning of generalization and average-case complexity and to formalize what this means for optimization algorithms. By bringing together experts in both fields, OPT 2021 will foster insightful discussions around these topics and more.
Workshop: ImageNet: Past, Present, and Future Mon 13 Dec 04:00 a.m.
Since its release in 2010, ImageNet has played an instrumental role in the development of deep learning architectures for computer vision, enabling neural networks to greatly outperform hand-crafted visual representations. ImageNet also quickly became the go-to benchmark for model architectures and training techniques which eventually reach far beyond image classification. Today’s models are getting close to “solving” the benchmark. Models trained on ImageNet have been used as strong initialization for numerous downstream tasks. The ImageNet dataset has even been used for tasks going way beyond its initial purpose of training classification model. It has been leveraged and reinvented for tasks such as few-shot learning, self-supervised learning and semi-supervised learning. Interesting re-creation of the ImageNet benchmark enables the evaluation of novel challenges like robustness, bias, or concept generalization. More accurate labels have been provided. About 10 years later, ImageNet symbolizes a decade of staggering advances in computer vision, deep learning, and artificial intelligence.
We believe now is a good time to discuss what’s next: Did we solve ImageNet? What are the main lessons learnt thanks to this benchmark? What should the next generation of ImageNet-like benchmarks encompass? Is language supervision a promising alternative? How can we reflect on the diverse requirements for good datasets and models, such as fairness, privacy, security, generalization, scale, and efficiency?
The pre-registration workshop: an alternative publication model for machine learning research Mon 13 Dec 04:00 a.m.
Machine learning research has benefited considerably from the adoption of standardised public benchmarks. While the importance of these benchmarks is undisputed, we argue against the current incentive system and its heavy reliance upon performance as a proxy for scientific progress. The status quo incentivises researchers to “beat the state of the art”, potentially at the expense of deep scientific understanding and rigorous experimental design. Since typically only positive results are rewarded, the negative results inevitably encountered during research are often omitted, allowing many other groups to unknowingly and wastefully repeat these negative findings.
Pre-registration is a publishing and reviewing model that aims to address these issues by changing the incentive system. A pre-registered paper is a regular paper that is submitted for peer-review without any experimental results, describing instead an experimental protocol to be followed after the paper is accepted. This implies that it is important for the authors to make compelling arguments from theory or past published evidence. As for reviewers, they must assess these arguments together with the quality of the experimental design, rather than comparing numeric results. While pre-registration has been highly adopted in fields such as medicine and psychology, there is little such experience inthe machine learning community. In this workshop, we propose to conduct a full pre-registration review-cycle for machine learning. Our proposal follows an initial small-scale trial of pre-registration in computer vision (Henriques et al., 2019) and builds on a successful pilot study in pre-registration at NeurIPS 2020 (Bertinetto et al., 2020). We have already received a number of requests to repeat the workshop, indicating strong community interest.
I (Still) Can't Believe It's Not Better: A workshop for “beautiful” ideas that "should" have worked Mon 13 Dec 04:50 a.m.
Beautiful ideas have shaped scientific progress throughout history. As Paul Dirac said, “If one is working from the point of view of getting beauty in one's equations, (…), one is on a sure line of progress.” However, beautiful ideas are often overlooked in a research environment that heavily emphasizes state-of-the-art (SOTA) results, where the worth of scientific works is defined by their immediate utility and quantitative superiority instead of their creativity, diversity, and elegance. This workshop will explore gaps between the form and function (or, the intrinsic and extrinsic value) of ideas in ML and AI research. We will explore that disconnect by asking researchers to submit their “beautiful” ideas that don’t (yet) “work”. We will ask them to explain why their idea has intrinsic value, and hypothesize why it hasn’t (yet) shown its extrinsic value. In doing so, we will create a space for researchers to help each other get their “beautiful” ideas “working”.
Workshop: Efficient Natural Language and Speech Processing (Models, Training, and Inference) Mon 13 Dec 05:00 a.m.
This workshop aims at introducing some fundamental problems in the field of natural language and speech processing which can be of interest to the general machine learning and deep learning community to improve the efficiency of the models, their training and inference. The workshop program offers an interactive platform for gathering experts and talents from academia and industry through different invited keynote talks, panel discussions, paper submissions, reviews, posters, oral presentations and a mentorship program.
This will provide an opportunity to discuss and learn from each other, exchange ideas, build connections, and brainstorm on potential solutions and future collaborations. The topics of this workshop can be of interest for people working on general machine learning, deep learning, optimization, theory and NLP & Speech applications.
Call for Papers
We encourage the NeurIPS community to submit their solutions, ideas, and ongoing work concerning data, model, training, and inference efficiency for NLP and speech processing. The scope of this workshop includes, but not limited to, the following topics.
(For more details please visit the Workshop Homepage.)
- Efficient Pre-Training and Fine-Tuning
- Model Compression
- Efficient Training
- Data Efficiency
- Edge Intelligence
Important Dates:
- Submission Deadline: September 18, 2021 (AOE)
- Acceptance Notification: October 22, 2021
- Camera-Ready Submission: November 1, 2021
- Workshop Date: December 13, 2021
Workshop: AI for Science: Mind the Gaps Mon 13 Dec 05:00 a.m.
Machine learning (ML) has revolutionized a wide array of scientific disciplines, including chemistry, biology, physics, material science, neuroscience, earth science, cosmology, electronics, mechanical science. It has solved scientific challenges that were never solved before, e.g., predicting 3D protein structure, imaging black holes, automating drug discovery, and so on. Despite this promise, several critical gaps stifle algorithmic and scientific innovation in "AI for Science": (1) Unrealistic methodological assumptions or directions, (2) Overlooked scientific questions, (3) Limited exploration at the intersections of multiple disciplines, (4) Science of science, (5) Responsible use and development of AI for science.
However, very little work has been done to bridge these gaps, mainly because of the missing link between distinct scientific communities. While many workshops focus on AI for specific scientific disciplines, they are all concerned with the methodological advances within a single discipline (e.g., biology) and are thus unable to examine the crucial questions mentioned above. This workshop will fulfill this unmet need and facilitate community building; with hundreds of ML researchers beginning projects in this area, the workshop will bring them together to consolidate the fast-growing area of "AI for Science" into a recognized field.
Workshop: Machine Learning Meets Econometrics (MLECON) Mon 13 Dec 05:00 a.m.
The Machine Learning Meets Econometrics (MLECON) workshop will serve as an interface for researchers from machine learning and econometrics to understand challenges and recognize opportunities that arise from the synergy between these two disciplines as well as to exchange new ideas that will help propel the fields. Our one-day workshop will consist of invited talks from world-renowned experts, shorter talks from contributed authors, a Gather.Town poster session, and an interdisciplinary panel discussion. To encourage cross-over discussion among those publishing in different venues, the topic of our panel discussion will be “Machine Learning in Social Systems: Challenges and Opportunities from Program Evaluation”. It was designed to highlight the complexity of evaluating social and economic programs as well as shortcomings of current approaches in machine learning and opportunities for methodological innovation. These challenges include more complex environments (markets, equilibrium, temporal considerations) and behavior (heterogeneity, delayed effects, unobserved confounders, strategic response). Our team of organizers and program committees is diverse in terms of gender, race, affiliations, country of origin, disciplinary background, and seniority levels. We aim to convene a broad variety of viewpoints on methodological axes (nonparametrics, machine learning, econometrics) as well as areas of application. Our invited speakers and panelists are leading experts in their respective fields and span far beyond the core NeurIPS community. Lastly, we expect participants with diverse backgrounds from various sub-communities of machine learning and econometrics (e.g., non- and semi-parametric econometrics, applied econometrics, reinforcement learning, kernel methods, deep learning, micro- and macro-economics) among other related communities.
Workshop: Optimal Transport and Machine Learning Mon 13 Dec 05:00 a.m.
Over the last few years, optimal transport (OT) has quickly become a central topic in machine learning. OT is now routinely used in many areas of ML, ranging from the theoretical use of OT flow for controlling learning algorithms to the inference of high-dimensional cell trajectories in genomics. The Optimal Transport and Machine Learning (OTML) workshop series (in '14, '17, '19) has been instrumental in shaping this research thread. For this new installment of OTML, we aim even bigger by hosting an exceptional keynote speaker, Alessio Figalli, who received the 2018 Fields Medal for his breakthroughs in the analysis of the regularity properties of OT. OTML will be a unique opportunity for cross-fertilization between recent advances in pure mathematics and challenging high-dimensional learning problems.
Workshop: New Frontiers in Federated Learning: Privacy, Fairness, Robustness, Personalization and Data Ownership Mon 13 Dec 05:30 a.m.
Federated Learning (FL) has recently emerged as the de facto framework for distributed machine learning (ML) that preserves the privacy of data, especially in the proliferation of mobile and edge devices with their increasing capacity for storage and computation. To fully utilize the vast amount of geographically distributed, diverse and privately owned data that is stored across these devices, FL provides a platform on which local devices can build their own local models whose training processes can be synchronized via sharing differential parameter updates. This was done without exposing their private training data, which helps mitigate the risk of privacy violation, in light of recent policies such as the General Data Protection Regulation (GDPR). Such potential use of FL has since then led to an explosive attention from the ML community resulting in a vast, growing amount of both theoretical and empirical literature that push FL so close to being the new standard of ML as a democratized data analytic service.
Interestingly, as FL comes closer to being deployable in real-world scenarios, it also surfaces a growing set of challenges on trustworthiness, fairness, auditability, scalability, robustness, security, privacy preservation, decentralizability, data ownership and personalizability that are all becoming increasingly important in many interrelated aspects of our digitized society. Such challenges are particularly important in economic landscapes that do not have the presence of big tech corporations with big data and are instead driven by government agencies and institutions with valuable data locked up or small-to-medium enterprises & start-ups with limited data and little funding. With this forethought, the workshop envisions the establishment of an AI ecosystem that facilitates data and model sharing between data curators as well as interested parties in the data and models while protecting personal data ownership.
Poster Session: [ protected link dropped ]
Workshop: Metacognition in the Age of AI: Challenges and Opportunities Mon 13 Dec 05:45 a.m.
Recent progress in artificial intelligence has transformed the way we live, work, and interact. Machines are mastering complex games and are learning increasingly challenging manipulation skills. Yet where are the robot agents that work for, with, and alongside us? These recent successes rely heavily on the ability to learn at scale, often within the confines of a virtual environment. This presents significant challenges for embodied systems acting and interacting in the real world. In contrast, we require our robots and algorithms to operate robustly in real-time, to learn from a limited amount of data, to take mission and sometimes safety-critical decisions, and increasingly even to display a knack for creative problem solving. Achieving this goal will require artificial agents to be able to assess - or introspect - their own competencies and their understanding of the world. Faced with similar complexity, there are a number of cognitive mechanisms which allow humans to act and interact successfully in the real world. Our ability to assess the quality of our own thinking - that is, our capacity for metacognition - plays a central role in this. We posit that recent advances in machine learning have, for the first time, enabled the effective implementation and exploitation of similar processes in artificial intelligence. This workshop brings together experts from psychology and cognitive science with cutting-edge research in machine learning, robotics, representation learning and related disciplines, with the ambitious aim of re-assessing how models of intelligence and metacognition can be leveraged in artificial agents given the potency of the toolset now available.
Workshop: Databases and AI (DBAI) Mon 13 Dec 05:50 a.m.
Relational data represents the vast majority of data present in the enterprise world. Yet none of the ML computations happens inside a relational database where data reside. Instead a lot of time is wasted in denormalizing the data and moving them outside of the databases in order to train models. Relational learning, which takes advantage of relational data structure, has been a 20 year old research area, but it hasn’t been connected with relational database systems, despite the fact that relational databases are the natural space for storing relational data. Recent advances in database research have shown that it is possible to take advantage of the relational structure in data in order to accelerate ML algorithms. Research in relational algebra originating from the database community has shown that it is possible to further accelerate linear algebra operations. Probabilistic Programming has also been proposed as a framework for AI that can be realized in relational databases. Data programming, a mechanism for weak/self supervision is slowly migrating to the natural space of storing data, the database. At last, as models in deep learning grow, several systems are being developed for model management inside relational databases
Workshop: Meaning in Context: Pragmatic Communication in Humans and Machines Mon 13 Dec 05:55 a.m.
Pragmatics – the aspects of language use that involve reasoning about context and other agents’ goals and belief states – has traditionally been treated as the “wastebasket” of language research (Bar-Hillel 1971), posing a challenge for both cognitive theories and artificial intelligence systems. Ideas from theoretical linguistics have inspired computational applications, such as in referential expression generation (Krahmer and van Deemter, 2012) or computational models of dialogue and recognition of speech or dialogue acts (Bunt and Black, 2000; Jurafsky, 2006; Ginzburg and Fernández, 2010; Bunt, 2016). But only recently, powerful artificial models based on neural or subsymbolic architectures have come into focus that generate or interpret language in pragmatically sophisticated and potentially open-ended ways (Golland et al. 2010, Andreas and Klein 2016, Monroe et al. 2017, Fried et al. 2018), building upon simultaneous advances in the cognitive science of pragmatics (Franke 2011, Frank and Goodman 2012). However, such models still fall short of human pragmatic reasoning in several important aspects. For example, existing approaches are often tailored to, or even trained to excel on, a specific pragmatic task (e.g., Mao et al. (2016) on discriminatory object description), leaving human-like task flexibility unaccounted for. It also remains largely underexplored how pragmatics connects to domain-general reasoning, how it may be efficiently implemented, and how it may arise over the course of learning and evolution. In this workshop, we aim to bring together researchers from Cognitive Science, Linguistics, and Machine Learning to think critically about the next generation of artificial pragmatic agents and theories of human pragmatic reasoning.
Workshop: Machine Learning and the Physical Sciences Mon 13 Dec 06:00 a.m.
The "Machine Learning and the Physical Sciences" workshop aims to provide a cutting-edge venue for research at the interface of machine learning (ML) and the physical sciences. This interface spans (1) applications of ML in physical sciences (“ML for physics”) and (2) developments in ML motivated by physical insights (“physics for ML”).
ML methods have had great success in learning complex representations of data that enable novel modeling and data processing approaches in many scientific disciplines. Physical sciences span problems and challenges at all scales in the universe: from finding exoplanets in trillions of sky pixels, to finding ML inspired solutions to the quantum many-body problem, to detecting anomalies in event streams from the Large Hadron Collider, to predicting how extreme weather events will vary with climate change. Tackling a number of associated data-intensive tasks including, but not limited to, segmentation, 3D computer vision, sequence modeling, causal reasoning, generative modeling, and efficient probabilistic inference are critical for furthering scientific discovery. In addition to using ML models for scientific discovery, tools and insights from the physical sciences are increasingly brought to the study of ML models.
By bringing together ML researchers and physical scientists who apply and study ML, we expect to strengthen the interdisciplinary dialogue, introduce exciting new open problems to the broader community, and stimulate the production of new approaches to solving challenging open problems in the sciences. Invited talks from leading individuals in both communities will cover the state-of-the-art techniques and set the stage for this workshop, which will also include contributed talks selected from submissions. The workshop will also feature an expert panel discussion on “Physics for ML" and a breakout session dedicated to community building will serve to foster dialogue between physical science and ML research communities.
Workshop: Machine Learning in Structural Biology Mon 13 Dec 06:00 a.m.
Structural biology, the study of proteins and other biomolecules through their 3D structures, is a field on the cusp of transformation. While measuring and interpreting biomolecular structures has traditionally been an expensive and difficult endeavor, recent machine-learning based modeling approaches have shown that it will become routine to predict and reason about structure at proteome scales with unprecedented atomic resolution. This broad liberation of 3D structure within bioscience and biomedicine will likely have transformative impacts on our ability to create effective medicines, to understand and engineer biology, and to design new molecular materials and machinery. Machine learning also shows great promise to continue to revolutionize many core technical problems in structural biology, including protein design, modeling protein dynamics, predicting higher order complexes, and integrating learning with experimental structure determination.
At this inflection point, we hope that the Machine Learning in Structural Biology (MLSB) workshop will help bring community and direction to this rising field. To achieve these goals, this workshop will bring together researchers from a unique and diverse set of domains, including core machine learning, computational biology, experimental structural biology, geometric deep learning, and natural language processing.
Differentiable Programming Workshop Mon 13 Dec 06:00 a.m.
Differentiable programming allows for automatically computing derivatives of functions within a high-level language. It has become increasingly popular within the machine learning (ML) community: differentiable programming has been used within backpropagation of neural networks, probabilistic programming, and Bayesian inference. Fundamentally, differentiable programming frameworks empower machine learning and its applications: the availability of efficient and composable automatic differentiation (AD) tools has led to advances in optimization, differentiable simulators, engineering, and science.
While AD tools have greatly increased the productivity of ML scientists and practitioners, many problems remain unsolved. Crucially, there is little communication between the broad group of AD users, the programming languages researchers, and the differentiable programming developers, resulting in them working in isolation. We propose a Differentiable Programming workshop as a forum to narrow the gaps between differentiable and probabilistic languages design, efficient automatic differentiation engines and higher-level applications of differentiable programming. We hope this workshop will harness a closer collaboration between language designers and domain scientists by bringing together a diverse part of the differentiable programming community including people working on core automatic differentiation tools, higher level frameworks that rely upon AD (such as probabilistic programming and differentiable simulators), and applications that use differentiable programs to solve scientific problems.
The explicit goals of the workshop are to:
1. Foster closer collaboration and synergies between the individual communities;
2. Evaluate the merits of differentiable design constructs and the impact they have on the algorithm design space and usability of the language;
3. Highlight differentiable techniques of individual domains, and the potential they hold for other fields.
Workshop: Shared Visual Representations in Human and Machine Intelligence Mon 13 Dec 06:45 a.m.
The goal of the 3rd Shared Visual Representations in Human and Machine Intelligence \textit{(SVRHM)} workshop is to disseminate relevant, parallel findings in the fields of computational neuroscience, psychology, and cognitive science that may inform modern machine learning. In the past few years, machine learning methods---especially deep neural networks---have widely permeated the vision science, cognitive science, and neuroscience communities. As a result, scientific modeling in these fields has greatly benefited, producing a swath of potentially critical new insights into the human mind. Since human performance remains the gold standard for many tasks, these cross-disciplinary insights and analytical tools may point towards solutions to many of the current problems that machine learning researchers face (\textit{e.g.,} adversarial attacks, compression, continual learning, and self-supervised learning). Thus we propose to invite leading cognitive scientists with strong computational backgrounds to disseminate their findings to the machine learning community with the hope of closing the loop by nourishing new ideas and creating cross-disciplinary collaborations. In particular, this year's version of the workshop will have a heavy focus on testing new inductive biases on novel datasets as we work on tasks that go beyond object recognition.
Workshop: Causal Inference & Machine Learning: Why now? Mon 13 Dec 07:00 a.m.
Machine Learning has been extremely successful throughout many critical areas, including computer vision, natural language processing, and game-playing. Still, a growing segment of the machine learning community recognizes that there are still fundamental pieces missing from the AI puzzle, among them causal inference.
This recognition comes from the observation that even though causality is a central component found throughout the sciences, engineering, and many other aspects of human cognition, explicit reference to causal relationships is largely missing in current learning systems. This entails a new goal of integrating causal inference and machine learning capabilities into the next generation of intelligent systems, thus paving the way towards higher levels of intelligence and human-centric AI. The synergy goes in both directions; causal inference benefitting from machine learning and the other way around. Current machine learning systems lack the ability to leverage the invariances imprinted by the underlying causal mechanisms towards reasoning about generalizability, explainability, interpretability, and robustness. Current causal inference methods, on the other hand, lack the ability to scale up to high-dimensional settings, where current machine learning systems excel.
The goal of this workshop is to bring together researchers from both camps to initiate principled discussions about the integration of causal reasoning and machine learning perspectives to help tackle the challenging AI tasks of the coming decades. We welcome researchers from all relevant disciplines, including but not limited to computer science, cognitive science, robotics, mathematics, statistics, physics, and philosophy.
Workshop: Human Centered AI Mon 13 Dec 07:00 a.m.
Human-Centered AI (HCAI) is an emerging discipline that aims to create AI systems that amplify [46,45] and augment [47] human abilities and preserve human control in order to make AI partnerships more productive, enjoyable, and fair [19]. Our workshop aims to bring together researchers and practitioners from the NeurIPS and HCI communities and others with convergent interests in HCAI.With an emphasis on diversity and discussion, we will explore research questions that stem from the increasingly wide-spread usage of machine learning algorithms across all areas of society, with a specific focus on understanding both technical and design requirements for HCAI systems, as well as how to evaluate the efficacy and effects of HCAI systems
Workshop: Machine Learning for Autonomous Driving Mon 13 Dec 07:50 a.m.
We propose a full-day workshop, called “Machine Learning for Autonomous Driving” (ML4AD), as a venue for machine learning (ML) researchers to discuss research problems concerning autonomous driving (AD). Our goal is to promote ML research, and its real-world impact, on self-driving technologies. Full self-driving capability (“Level 5”) is far from solved and extremely complex, beyond the capability of any one institution or company, necessitating larger-scale communication and collaboration, which we believe workshop formats help provide.
We propose a large-attendance talk format of approximately 500 attendees, including (1) a call for papers with poster sessions and spotlight presentations; (2) keynote talks to communicate the state-of-the-art; (3) panel debates to discuss future research directions; (4) a call for challenge to encourage interaction around a common benchmark task; (5) social breaks for newer researchers to network and meet others.
Workshop on Deep Learning and Inverse Problems Mon 13 Dec 07:55 a.m.
Learning-based methods, and in particular deep neural networks, have emerged as highly successful and universal tools for image and signal recovery and restoration. They achieve state-of-the-art results on tasks ranging from image denoising, image compression, and image reconstruction from few and noisy measurements. They are starting to be used in important imaging technologies, for example in GEs newest computational tomography scanners and in the newest generation of the iPhone.
The field has a range of theoretical and practical questions that remain unanswered, including questions about guarantees, robustness, architectural design, the role of learning, domain-specific applications, and more. This virtual workshop aims at bringing together theoreticians and practitioners in order to chart out recent advances and discuss new directions in deep learning-based approaches for solving inverse problems in the imaging sciences and beyond.
Workshop: CtrlGen: Controllable Generative Modeling in Language and Vision Mon 13 Dec 08:00 a.m.
Over the past few years, there has been an increased interest in the areas of language and image generation within the community. As generated texts by models like GPT-3 start to sound more fluid and natural, and generated images and videos by GAN models appear more realistic, researchers began focusing on qualitative properties of the generated content such as the ability to control its style and structure, or incorporate information from external sources into the output. Such aims are extremely important to make language and image generation useful for human-machine interaction and other real-world applications including machine co-creativity, entertainment, reducing biases or toxicity, and improving conversational agents and personal assistants.
Achieving these ambitious but important goals introduces challenges not only from NLP and Vision perspectives, but also ones that pertain to Machine Learning as a whole, which has witnessed a growing body of research in relevant domains such as interpretability, disentanglement, robustness, and representation learning. We believe that progress towards the realization of human-like language and image generation may benefit greatly from insights and progress in these and other ML areas.
In this workshop, we propose to bring together researchers from the NLP, Vision, and ML communities to discuss the current challenges and explore potential directions for controllable generation and improve its quality, correctness, and diversity. As excitement about language and image generation has significantly increased recently thanks to the advent and improvement of language models, Transformers, and GANs, we feel this is the opportune time to hold a new workshop about this subject. We hope CtrlGen will foster discussion and interaction across communities, and sprout fruitful cross-domain relations that open the door for enhanced controllability in language and image generation.
Workshop: Safe and Robust Control of Uncertain Systems Mon 13 Dec 08:00 a.m.
Control and decision systems are becoming a ubiquitous part of our daily lives, ranging from serving advertisements or recommendations on the internet to controlling autonomous physical systems such as industrial equipment or robots. While these systems have shown the potential for significantly improving quality of life and industrial efficiency, the impact of the decisions made by these systems can also cause significant damages. For example, an online retailer recommending dangerous products to children, a social media platform serving content which polarizes society, or a household robot/autonomous car which collides with surrounding humans can all cause significant direct harm to society. These undesirable behaviors not only can be dangerous, but also lead to significant inefficiencies when deploying learning-based agents in the real world. This motivates developing algorithms for learning-based control which can reason about uncertainty and constraints in the environment to explicitly avoid undesirable behaviors. We believe hosting a discussion on safety in learning-based control at NeurIPS 2021 would have far-reaching societal impacts by connecting researchers from a variety of disciplines including machine learning, control theory, AI safety, operations research, robotics, and formal methods.
Workshop: Machine Learning for Creativity and Design Mon 13 Dec 08:15 a.m.
Machine co-creativity continues to grow and attract a wider audience to machine learning. Generative models, for example, have enabled new types of media creation across language, images, and music--including recent advances such as CLIP, VQGAN, and DALL·E. This one-day workshop will broadly explore topics in the applications of machine learning to creativity and design, which includes:
State-of-the-art algorithms for the creation of new media. Machine learning models achieving state-of-the-art in traditional media creation tasks (e.g., image, audio, or video synthesis) that are also being used by the artist community will be showcased.
Artist accessibility of machine learning models. Researchers building the next generation of machine learning models for media creation will be challenged in understanding the accessibility needs of artists. Artists and Human Computer interaction / User Experience community members will be encouraged to engage in the conversation.
The sociocultural and political impact of these new models. With the increased popularity of generative machine learning models, we are witnessing these models start to impact our everyday surroundings, ranging from racial and gender bias in algorithms and datasets used for media creation to how new media manipulation tools may erode our collective trust in media content.
Artistic applications. We will hear directly from some of the artists who are adopting machine learning--including deep learning and reinforcement learning--as part of their own artistic process as well as showcasing their work.
The goal of this workshop is to bring together researchers and artists interested in exploring the intersection of human creativity and machine learning and foster collaboration between them, as well as promote the sharing of ideas, perspectives, new research, artwork, and artistic needs.
Discord invite --> https://bit.ly/3puzVuM
Workshop: Learning in Presence of Strategic Behavior Mon 13 Dec 08:50 a.m.
In recent years, machine learning has been called upon to solve increasingly more complex tasks and to regulate many aspects of our social, economic, and technological world. These applications include learning economic policies from data, prediction in financial markets, learning personalize models across population of users, and ranking qualified candidates for admission, hiring, and lending. These tasks take place in a complex social and economic context where the learners and objects of learning are often people or organizations that are impacted by the learning algorithm and, in return, can take actions that influence the learning process. Learning in this context calls for a new vision for machine learning and economics that aligns the incentives and interests of the learners and other parties and is robust to the evolving social and economic needs. This workshop explores a view of machine learning and economics that considers interactions of learning systems with a wide range of social and strategic behaviors. Examples of these problems include: multi-agent learning systems, welfare-aware machine learning, learning from strategic and economic data, learning as a behavioral model, and causal inference for learning impact of strategic choices.
Workshop: Deep Reinforcement Learning Mon 13 Dec 08:55 a.m.
In recent years, the use of deep neural networks as function approximators has enabled researchers to extend reinforcement learning techniques to solve increasingly complex control tasks. The emerging field of deep reinforcement learning has led to remarkable empirical results in rich and varied domains like robotics, strategy games, and multiagent interactions. This workshop will bring together researchers working at the intersection of deep learning and reinforcement learning, and it will help interested researchers outside of the field gain perspective about the current state of the art and potential directions for future contributions.
Workshop: ML For Systems Mon 13 Dec 08:55 a.m.
ML for Systems is an emerging research area that has shown promising results in the past few years. Recent work has shown that ML can be used to replace heuristics, solve complex optimization problems, and improve modeling and forecasting when applied in the context of computer systems.
As an emerging area, ML for Systems is still in the process of defining the common problems, frameworks and approaches to solving its problems, which requires venues that bring together researchers and practitioners from both the systems and machine learning communities. Past iterations of the workshops focused on providing such a venue and broke new ground on a broad range of emerging new directions in ML for Systems. We want to carry this momentum forward by encouraging the community to explore areas that have previously received less attention. Specifically, the workshop commits to highlighting works that also optimize for security and privacy, as opposed to metrics like speed and memory and use ML to optimize for energy usage and carbon impact. Additionally, this year we will encourage the development of shared methodology, tools, and frameworks.
For the first time since the inception of the workshop, we will organize a competition. This competition will showcase important systems problems, and challenges the ML community to test their methods and algorithms on these problems. Our competition tasks are designed to have a low barrier of entry that attracts newcomers as well as systems veterans.
This setup will allow attendees to meet with top researchers and domain experts, old and new, bridging cutting edge ML research with practical systems design. We hope that providing a prestigious venue for researchers from both fields to meet and interact will result in both fundamental ML research as well as real-world impact to computer systems design and implementation.
Third Workshop on AI for Humanitarian Assistance and Disaster Response Mon 13 Dec 09:00 a.m.
Natural disasters are one of the oldest threats to both individuals and the societies they co-exist in. As a result, humanity has ceaselessly sought way to provide assistance to people in need after disasters have struck. Further, natural disasters are but a single, extreme example of the many possible humanitarian crises. Disease outbreak, famine, and oppression against disadvantaged groups can pose even greater dangers to people that have less obvious solutions. In this proposed workshop, we seek to bring together the Artificial Intelligence (AI) and Humanitarian Assistance and Disaster Response (HADR) communities in order to bring AI to bear on real-world humanitarian crises. Through this workshop, we intend to establish meaningful dialogue between the communities.
By the end of the workshop, the NeurIPS research community can come to understand the practical challenges of aiding those who are experiencing crises, while the HADR community can understand the landscape that is the state of art and practice in AI. Through this, we seek to begin establishing a pipeline of transitioning the research created by the NeurIPS community to real-world humanitarian issues.
Workshop: Distribution shifts: connecting methods and applications (DistShift) Mon 13 Dec 09:00 a.m.
Distribution shifts---where a model is deployed on a data distribution different from what it was trained on---pose significant robustness challenges in real-world ML applications. Such shifts are often unavoidable in the wild and have been shown to substantially degrade model performance in applications such as biomedicine, wildlife conservation, sustainable development, robotics, education, and criminal justice. For example, models can systematically fail when tested on patients from different hospitals or people from different demographics. Despite the ubiquity of distribution shifts in ML applications, work on these types of real-world shifts is currently underrepresented in the ML research community, with prior work generally focusing instead on synthetic shifts. However, recent work has shown that models that are robust to one kind of shift need not be robust to another, underscoring the importance and urgency of studying the types of distribution shifts that arise in real-world ML deployments. With this workshop, we aim to facilitate deeper exchanges between domain experts in various ML application areas and more methods-oriented researchers, and ground the development of methods for characterizing and mitigating distribution shifts in real-world application contexts.