Aike AI frontier promotion (7.23)

LG - machine learning CV - Computer vision CL - Computing and language AS - Audio and voice RO - robot

Turn from love to a lovely life

Abstract ： Identification and characterization of rumor super disseminators on social media 、 Language model cascade 、 Human value construction in Recommendation System 、 Evaluation of generalization of transfer learning distribution 、 Summarize the influence of deviation on scaling 、 Instance aware observer network for distributed object segmentation 、 Teach qubits to sing 、 Neural bilateral filtering for point cloud top-down instance segmentation 、 Generalization of block based neural rendering

1、[SI] Identification and characterization of misinformation superspreaders on social media

M R. DeVerna, R Aiyappa, D Pacheco, J Bryden, F Menczer

[Indiana University]

Identification and characterization of rumor super disseminators on social media . The digital information ecosystem in the world is still associated with rumors ( error message ) Fight against the spread of . Previous work has shown that , Users who continue to disseminate a large number of low confidence content —— The so-called super communicator —— Is the center of this problem . This paper quantitatively confirms this hypothesis , A simple indicator is introduced to predict the top rumor super disseminators in the coming months . A qualitative review was conducted , To identify the characteristics of the most productive super communicators , Analyze their sharing behavior . Super communicators include scholars with a large number of fans 、 Low reputation media organizations 、 Personal accounts related to these media institutions , And a series of influential people . They are mainly political , Use more toxic language than typical users who share error messages . This paper also finds worrying evidence , indicate Twitter May have overlooked prominent super communicators . It is hoped that this work can promote public understanding of bad actors , And promote the adoption of measures , Reduce their negative impact on healthy digital discourse .

The world’s digital information ecosystem continues to struggle with the spread of misinformation. Prior work has suggested that users who consistently disseminate a disproportionate amount of lowcredibility content — so-called superspreaders — are at the center of this problem. We quantitatively confirm this hypothesis and introduce simple metrics to predict the top misinformation superspreaders several months into the future. We then conduct a qualitative review to characterize the most prolific superspreaders and analyze their sharing behaviors. Superspreaders include pundits with large followings, low-credibility media outlets, personal accounts affiliated with those media outlets, and a range of influencers. They are primarily political in nature and use more toxic language than the typical user sharing misinformation. We also find concerning evidence suggesting that Twitter may be overlooking prominent superspreaders. We hope this work will further public understanding of bad actors and promote steps to mitigate their negative impacts on healthy digital discourse.

https://arxiv.org/abs/2207.09524

2、[CL] Language Model Cascades

D Dohan, W Xu, A Lewkowycz, J Austin, D Bieber, R G Lopes, Y Wu, H Michalewski, R A. Saurous, J Sohl-dickstein, K Murphy, C Sutton

[Google Research & Alphabet]

Language model cascade . The hint model has shown impressive learning ability with few samples . Repeated interaction with a single model during testing , Or combine multiple models , Further expand the capacity . These combinations are probabilistic models , It can be expressed in the graph model language of random variables , Its value is a complex data type , Such as a string . Cases with control flow and dynamic structure require techniques from probabilistic programming , These technologies allow different model structures and reasoning strategies to be implemented in a unified language . This paper formalizes several existing technologies from this perspective , Include scratchpads/ Thinking chain 、 Validator 、STaR、 Select reasoning and tool use . The generated program is called language model cascade .

Prompted models have demonstrated impressive few-shot learning abilities. Repeated interactions at test-time with a single model, or the composition of multiple models together, further expands capabilities. These compositions are probabilistic models, and may be expressed in the language of graphical models with random variables whose values are complex data types such as strings. Cases with control flow and dynamic structure require techniques from probabilistic programming, which allow implementing disparate model structures and inference strategies in a unified language. We formalize several existing techniques from this perspective, including scratchpads / chain of thought, verifiers, STaR, selection-inference, and tool use. We refer to the resulting programs as language model cascades.

https://arxiv.org/abs/2207.10342

3、[IR] Building Human Values into Recommender Systems: An Interdisciplinary Synthesis

J Stray, A Halevy, P Assar, D Hadfield-Menell, C Boutilier, A Ashar...

[UC Berkeley & Meta AI & MIT & Google Research & Spotify Inc...]

Human value construction in Recommendation System ： Interdisciplinary comprehensive report . Recommender systems are chosen from many of the world's largest platforms and Applications 、 Algorithms for filtering and personalizing content . therefore , Their positive and negative effects on individuals and society have been widely discussed and studied . The primary problem of this paper is how to ensure that recommendation systems can realize the personal and social value they serve . Solving this problem in a principled way requires technical knowledge of recommender design and operation , It also depends on insights from different fields , Including Social Sciences 、 ethics 、 economics 、 psychology 、 Policies and laws . This paper is a multidisciplinary effort , It integrates theory and practice from different angles , The aim is to provide a common language , Clarify current design methods , And identify open questions . It is not a comprehensive survey of this big space , But a group of highlights determined by different groups of authors of this article . This paper collects a set of values that are most relevant to recommendation systems in different fields , Then from the current industry practice 、 measurement 、 They are studied from the perspective of product design and policy methods . Important open-ended issues include the multi stakeholder process of defining values and resolving trade-offs 、 Better value driven measurement 、 People use recommenders to control 、 Non behavioral algorithm feedback 、 Optimization of long-term results 、 Causal inference of recommender effect 、 academic - Industrial research cooperation and interdisciplinary policy-making .

Recommender systems are the algorithms which select, filter, and personalize content across many of the world’s largest platforms and apps. As such, their positive and negative effects on individuals and on societies have been extensively theorized and studied. Our overarching question is how to ensure that recommender systems enact the values of the individuals and societies that they serve. Addressing this question in a principled fashion requires technical knowledge of recommender design and operation, and also critically depends on insights from diverse fields including social science, ethics, economics, psychology, policy and law. This paper is a multidisciplinary effort to synthesize theory and practice from different perspectives, with the goal of providing a shared language, articulating current design approaches, and identifying open problems. It is not a comprehensive survey of this large space, but a set of highlights identified by our diverse author cohort. We collect a set of values that seem most relevant to recommender systems operating across different domains, then examine them from the perspectives of current industry practice, measurement, product design, and policy approaches. Important open problems include multi-stakeholder processes for defining values and resolving trade-offs, better values-driven measurements, recommender controls that people use, non-behavioral algorithmic feedback, optimization for long-term outcomes, causal inference of recommender effects, academic-industry research collaborations, and interdisciplinary policy-making.

https://arxiv.org/abs/2207.10192

4、[LG] Assaying Out-Of-Distribution Generalization in Transfer Learning

F Wenzel, A Dittadi, P V Gehler, C Simon-Gabriel, M Horn, D Zietlow, D Kernert, C Russell...

[AWS Tübingen & Technical University of Denmark]

Evaluation of generalization of transfer learning distribution . Because the generalization of distribution is a common problem , Various agent goals have been studied in different projects ( for example , calibration 、 Antagonistic robustness 、 Algorithm corruption 、 Cross drift invariance ）, This leads to different suggestions . Although they have the same wishes and goals , But these methods have never been tested on real data under the same experimental conditions . This paper takes a unified view of the previous work , It emphasizes the information differences solved through experience , It also provides suggestions on how to measure the robustness of a model and how to improve it . So , This article collects 172 Publicly available data set pairs , Used for training and evaluation of accuracy outside the distribution 、 Calibration error 、 Counter attack 、 Environmental invariance and synthetic decay . For more than 31000 Networks have been fine tuned , These networks come from nine different architectures , Under the setting of multiple samples and less samples . The results confirm that , The accuracy inside and outside the distribution tends to increase jointly , Their relationship depends largely on the data set , And in general, it is more subtle and complex than previously assumed by smaller studies .

Since out-of-distribution generalization is a generally ill-posed problem, various proxy targets (e.g., calibration, adversarial robustness, algorithmic corruptions, invariance across shifts) were studied across different research programs resulting in different recommendations. While sharing the same aspirational goal, these approaches have never been tested under the same experimental conditions on real data. In this paper, we take a unified view of previous work, highlighting message discrepancies that we address empirically, and providing recommendations on how to measure the robustness of a model and how to improve it. To this end, we collect 172 publicly available dataset pairs for training and out-of-distribution evaluation of accuracy, calibration error, adversarial attacks, environment invariance, and synthetic corruptions. We fine-tune over 31k networks, from nine different architectures in the manyand few-shot setting. Our findings confirm that inand out-of-distribution accuracies tend to increase jointly, but show that their relation is largely dataset-dependent, and in general more nuanced and more complex than posited by previous, smaller scale studies.

https://arxiv.org/abs/2207.09239

5、[LG] Scaling Laws vs Model Architectures: How does Inductive Bias Influence Scaling?

Y Tay, M Dehghani, S Abnar, H W Chung, W Fedus, J Rao, S Narang, V Q. Tran, D Yogatama, D Metzler

[Google Research & DeepMind]

Scaling law and model architecture ： Summarize the influence of deviation on scaling . People are right. Transformer The scaling properties of the model are of great interest . However , In studying the influence of different inductive deviations and the scaling characteristics of the model structure , Not much work has been done . Whether the model architecture has different extensibility ？ If so , Summarize how deviation affects expansion behavior ？ How does this affect upstream ( Preliminary training ) And downstream ( transfer )？ This paper systematically studies the extension behavior of ten different model architectures , Such as Transformer、Switch Transformer、Universal Transformer、 Dynamic convolution 、Performer And recently proposed MLP-Mixer. Through extensive experiments , This article shows that ：（1） When expanding , Structure is indeed an important consideration ;（2） The best model will fluctuate on different scales . This article believes that , The findings in this work have a significant impact on the way the current community assessment model is structured .

There have been a lot of interest in the scaling properties of Transformer models (Kaplan et al., 2020). However, not much has been done on the front of investigating the effect of scaling properties of different inductive biases and model architectures. Do model architectures scale differently? If so, how does inductive bias affect scaling behaviour? How does this influence upstream (pretraining) and downstream (transfer)? This paper conducts a systematic study of scaling behaviour of ten diverse model architectures such as Transformers, Switch Transformers, Universal Transformers, Dynamic convolutions, Performers, and recently proposed MLP-Mixers. Via extensive experiments, we show that (1) architecture is an indeed an important consideration when performing scaling and (2) the best performing model can fluctuate at different scales. We believe that the findings outlined in this work has significant implications to how model architectures are currently evaluated in the community.

https://arxiv.org/abs/2207.10551

Several other papers worthy of attention ：

[CV] Instance-Aware Observer Network for Out-of-Distribution Object Segmentation

Instance aware observer network for distributed object segmentation

V Besnier, A Bursuc, D Picard, A Briot

[Valeo & Univ Gustave Eiffel]

https://arxiv.org/abs/2207.08782

[AI] Teaching Qubits to Sing: Mission Impossible?

Teach qubits to sing ： Impossible Mission ？

E R Miranda, B N. Siegelwax

[University of Plymouth]

https://arxiv.org/abs/2207.08225

[CV] NeuralBF: Neural Bilateral Filtering for Top-down Instance Segmentation on Point Clouds

NeuralBF： Neural bilateral filtering for point cloud top-down instance segmentation
W Sun, D Rebain, R Liao, V Tankovich, S Yazdani, K M Yi, A Tagliasacchi

[University of British Columbia & Google Research]

https://arxiv.org/abs/2207.09978