Model Inversion (MI), in which an adversary abuses access to a trained Machine Learning (ML) model attempting to infer sensitive information about its original training data, has attracted increasing research attention. During MI, the trained model under attack (MUA) is usually frozen and used to guide the training of a generator, such as a Generative Adversarial Network (GAN), to reconstruct the distribution of the original training data of that model. This might cause leakage of original training samples, and if successful, the privacy of dataset subjects will be at risk if the training data contains Personally Identifiable Information (PII). Therefore, an in-depth investigation of the potentials of MI techniques is crucial for the development of corresponding defense techniques. High-quality reconstruction of training data based on a single model is challenging. However, existing MI literature does not explore targeting multiple models jointly, which may provide additional information and diverse perspectives to the adversary.

We propose the ensemble inversion technique that estimates the distribution of original training data by training a generator constrained by an ensemble (or set) of trained models with shared subjects or entities. This technique leads to noticeable improvements of the quality of the generated samples with distinguishable features of the dataset entities compared to MI of a single ML model. We achieve high quality results without any dataset and show how utilizing an auxiliary dataset that's similar to the presumed training data improves the results. The impact of model diversity in the ensemble is thoroughly investigated and additional constraints are utilized to encourage sharp predictions and high activations for the reconstructed samples, leading to more accurate reconstruction of training images.

WACV

Reconstructing Training Data from Diverse ML Models by Ensemble Inversion

Spurious correlations allow flexible models to predict well during training but poorly on related test populations. Recent work has shown that models that satisfy particular independencies involving correlation-inducing nuisance variables have guarantees on their test performance. Enforcing such independencies requires nuisances to be observed during training. However, nuisances, such as demographics or image background labels, are often missing. Enforcing independence on just the observed data does not imply independence on the entire population. Here we derive MMD estimators used for invariance objectives under missing nuisances. On simulations and clinical data, optimizing through these estimates achieves test performance similar to using estimators that make use of the full data.

NeurIPS Workshop

Learning Invariant Representations with Missing Data

Providing new features—while preserving user privacy—requires techniques for learning from private and anonymized user feedback. To learn quickly and accurately, we develop and employ statistical learning algorithms that help us overcome multiple challenges that arise from sampling noise, applications of differential privacy, and delays that may be present in the data. These algorithms enable teams at Apple to measure and understand which user experiences are the best. This understanding leads to continual improvements across Apple's products and services to drive better experiences. We provide aspects of this understanding to the Apple developer community through features such as [product page optimization](https://developer.apple.com/app-store/product-page-optimization/).  

Interpretable Adaptive Optimization

Modeling the world can benefit robot learning by providing a rich training signal for shaping an agent's latent state space. However, learning world models in unconstrained environments over high-dimensional observation spaces such as images is challenging. One source of difficulty is the presence of irrelevant but hard-to-model background distractions, and unimportant visual details of task-relevant entities. We address this issue by learning a recurrent latent dynamics model which contrastively predicts the next observation. This simple model leads to surprisingly robust robotic control even with simultaneous camera, background, and color distractions. We outperform alternatives such as bisimulation methods which impose state-similarity measures derived from divergence in future reward or future optimal actions. We obtain state-of-the-art results on the Distracting Control Suite, a challenging benchmark for pixel-based robotic control.

Robust Robotic Control from Pixels Using Contrastive Recurrent State-Space Models

Apple machine learning teams are engaged in state of the art research in machine learning and artificial intelligence. Learn about the latest advancements.

Overview

Residents will have the opportunity to attend machine learning and AI courses, learn from an Apple mentor closely involved in their program, collaborate with fellow residents, gain hands-on experience working on high-impact projects, publish in premier academic conferences, and partner with Apple teams across hardware, software, and services.

Applications for the 2022 AI/ML residency program are now closed. Please check back next year for more information about our 2023 applications.

<div class="callout">

Discover other opportunities for researchers, students, and developers in Machine Learning at Apple [here](https://machinelearning.apple.com/work-with-us)

Explore advancements in Machine Learning research at Apple [here](https://machinelearning.apple.com/research/) 

</div>

<p>The AI/ML Residency Program invites experts in various fields to apply their own domain expertise to innovate and build revolutionary machine learning and AI-powered products and experiences. As AI-based solutions spread across disciplines, the need for domain experts to understand machine learning and apply their expertise in ML settings grows. The program aims to invest in the resident’s technical and theoretical machine learning development to help advance their professional careers.</p>

The AI/ML Residency Program


 ## Accepted Papers

 ### Conference Accepted Papers

### [ARKitScenes - A Diverse Real-World Dataset For 3D Indoor Scene Understanding Using Mobile RGB-D Data](/research/arkitscenes)

*Gilad Baruch, Zhuoyuan Chen, Afshin Dehghan, Tal Dimry, Yuri Feigin, Peter Fu, Thomas Gebauer, Brandon Joffe, Daniel Kurz, Arik Schwartz, Elad Shulman*

### [BatchQuant: Quantized-for-all Architecture Search with Robust Quantizer](/research/batchquant)

*Haoping Bai, Meng Cao, Ping Huang, Jiulong Shan*

### [Individual Privacy Accounting via a Renyi Filter](/research/individual-privacy-accounting)

*Vitaly Feldman, Tijana Zrnic*

### [Probabilistic Attention for Interactive Segmentation](/research/probabilistic-attention-for-interactive-segmentation)

*Prasad Gabbur, Manjot Bilkhu, Javier Movellan*

This paper was accepted to the main session at NeurIPS as a spotlight paper.

### [RIM: Reliable Influence-based Active Learning on Graphs](/research/rim)  

*Wentao Zhang, Yexin Wang, Zhenbang You, Meng Cao, Ping Huang, Jiulong Shan, Zhi Yang, Bin Cui*

This paper was accepted to the main session at NeurIPS as a spotlight paper.

### Workshop Accepted Papers

### [Challenges of Adversarial Image Augmentations](/research/challenges-adversarial-image)

*Arno Blaas, Xavier Suau Cuadros, Jason Ramapuram, Luca Zappella, Nicholas Apostoloff* 

### [Do Self-Supervised and Supervised Methods Learn Similar Visual Representations?](/research/self-supervised-supervised-methods)

*Tom Grigg, Dan Busbridge, Jason Ramapuram, Russ Webb* 

### [Enforcing Fairness in Private Federated Learning via the Modified Method of Differential Multipliers](/research/enforcing-fairness)

*Borja Rodríguez-Gálvez, Filip Granqvist, Rogier van Dalen, Matt Seigel*

### [Learning Invariant Representations with Missing Data](/research/learning-invariant-representations)

*Mark Goldstein, Jörn-Henrik Jacobsen, Olina Chau, Adriel Saporta, Aahlad Puli, Rajesh Ranganath, Andrew C. Miller*

### [Robust Robotic Control from Pixels Using Contrastive Recurrent State-Space Models](/research/contrastive-recurrent-state-space-models) 

*Nitish Srivastava, Walter Talbott, Martin Bertran Lopez, Shuangfei Zhai, Josh Susskind* 

### [Self-supervised Semi-supervised Learning for Data Labeling and Quality Evaluation](/research/self-supervised-semi-supervised-learning) 

*Haoping Bai, Meng Cao, Ping Huang, Jiulong Shan* 

### [Stochastic Contrastive Learning](/research/stochastic-contrastive-learning) 

*Jason Ramapuram, Dan Busbridge, Xavier Suau, Russ Webb* 

## Talks and Workshops

“Meet Apple” is an opportunity to learn more about our machine learning teams, working at Apple, and how to apply to full-time positions. This talk will be held virtually on December 8 at 1:15 pm PST. 

Apple will be hosting a panel on internships, where attendees can learn more about internship opportunities across our machine learning teams. It will be held virtually on December 8 at 2:30 pm PST.

Nick Foti is a co-organizer of the [Your Model is Wrong: Robustness and Misspecification in Probabilistic Modeling](https://sites.google.com/view/robustbayes-neurips21/home) workshop at NeurIPS, which will take place on December 14.

Yahav Bechavod and Tijana Zrnic, two of our Apple Scholars in AI/ML, have co-organized the [StratML](https://www.strategic-ml.com/) workshop at NeurIPS that will take place on December 14. While Apple is not a sponsor or organizer of the StratML workshop, we are excited to amplify the outstanding work and contributions of students who are Apple Scholars.

All registered NeurIPS attendees are invited to attend each event. Check out our [agenda](https://apple.ent.box.com/s/qmttmd62m8hlg66nnbtcs9cspwq7ps3v) for more information on how to join.

## Affinity Events

Apple is a sponsor of [LatinX in AI](https://www.latinxinai.org/neurips-2021-about), who will be holding a workshop at NeurIPS on December 7. Miguel Angel Bautista Martin will be participating in the mentoring hour that will take place during the workshop as a mentor.

Apple is also a sponsor of [Women in Machine Learning](https://sites.google.com/view/wiml2021), whose workshop will take place December 8 to 10. Lizi Ottens will be speaking about how our on-device machine learning powers intelligent experiences in Apple products.

In addition to LatinX in AI and Women in Machine Learning, Apple is a sponsor of [Queer in AI](https://sites.google.com/view/queer-in-ai/neurips-2021), who will be holding a workshop December 7 to 8.  

[Black in AI](https://blackinai.github.io/#/workshop/bai2021) will be hosting a workshop on December 10, and all registered NeurIPS attendees are encouraged to attend.

Learn more about Apple’s company-wide efforts in [inclusion and diversity](https://www.apple.com/diversity/). 

## Acknowledgements

Samy Bengio is a NeurIPS 2021 Board Member, and one of the Ethics Review Chairs of the NeurIPS 2021 Organizing Committee.

<div class="callout">

Let's innovate together. Build amazing machine-learned experiences with Apple. Discover opportunities for researchers, students, and developers by visiting our [Work With Us page.](/work-with-us)

</div>

<p>The Neural Information Processing Systems (NeurIPS) conference will be held virtually from December 6 - 14. NeurIPS is a global conference focused on fostering the exchange of research on neural information processing systems in their biological, technological, mathematical, and theoretical aspects.</p>

Machine Learning
Research at Apple.

Recent research.

Reconstructing Training Data from Diverse ML Models by Ensemble Inversion

Learning Invariant Representations with Missing Data

Interpretable Adaptive Optimization

Robust Robotic Control from Pixels Using Contrastive Recurrent State-Space Models

Updates and events.

The AI/ML Residency Program

Apple at NeurIPS 2021

Discover opportunities in Machine Learning.

Machine LearningResearch at Apple.

Recent research.

Reconstructing Training Data from Diverse ML Models by Ensemble Inversion

Learning Invariant Representations with Missing Data

Interpretable Adaptive Optimization

Robust Robotic Control from Pixels Using Contrastive Recurrent State-Space Models

Updates and events.

The AI/ML Residency Program

Apple at NeurIPS 2021

Discover opportunities in Machine Learning.

Machine Learning
Research at Apple.