A growing number of consumer devices, including smart speakers, headphones, and watches, use speech as the primary means of user input. As a result, voice trigger detection systems—a mechanism that uses voice recognition technology to control access to a particular device or feature—have become an important component of the user interaction pipeline as they signal the start of an interaction between the user and a device. Since these systems are deployed entirely on-device, several considerations inform their design, like privacy, latency, accuracy, and power consumption.

Voice Trigger System for Siri

The Photos app learns about significant people, places, and events based on the user’s library, and then presents Memories: curated collections of photos and videos set to music. The key photo for a Memory is influenced by the popularity of iconic scenes learned from iOS users—with DP assurance.

<Figure id="figure1" title="Figure 1" className="inline" caption="Figure 1: This feature automatically picks one photo from the Memory as the key image. The algorithm is powered, among other signals, by iconic scenes crowd-learned with differential privacy.">
<Video src="https://mlr.cdn-apple.com/video/Fig_1_screens_10bd45bc93.mp4" autoPlay loop />
</Figure>

We prioritized three key aspects of the machine learning research powering this feature: 

1. How to accommodate data changes given its sensitivity
2. How to navigate the tradeoffs between local and central differential privacy
3. How to accommodate non-uniform data density

**Data changes.** Identifying iconic scenes—like popular locations such as Central Park in New York City or the Golden Gate Bridge in San Francisco—and the type of photos people take at those locations allows for better key photo selection. Even so, we want to make Memories a compelling feature for device owners no matter where they are located. In addition, iconic scenes can change over time or depending on the season. For example, if a new dog park opens near the Golden Gate Bridge, people might start taking more photos of dogs in that location. Or, if it is winter, people might be taking pictures of the ice skating rink in Central Park. Identifying a new category of photos in a popular location is a perfect task for a data-driven solution. To preserve user privacy, we want to keep information about where people go and what they see there on their device. 

**Local versus central differential privacy.** DP was designed to learn statistics about data with strong assurances of not leaking personally identifiable information. With these considerations in mind, we started with local DP solutions, as seen in the paper [Learning with Privacy at Scale,](https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf) that implement sufficient protections in the operating system so that researchers could verify our claims by reverse-engineering the code. However, we soon discovered that this approach was limited since local DP requires a significant amount of noise to be added, allowing us to discover only the most prominent signal. We needed a better solution that could provide a transparent and verifiable DP assurance while also improving the utility of the learned histograms.

**Nonuniform data density.** Data gathered across the world is not uniformly distributed, so the privacy challenges differ across regions. For example, suppose many users take photos in a particular location. This makes learning detailed statistics with high precision and privacy assurances easier and provides more useful results. If fewer users visit other locations and take fewer photos, we have less data to work from. This makes it more difficult to glean a meaningful signal without jeopardizing user privacy. In high-density areas, we could learn better precision (with the same privacy assurance) because there is a bigger crowd to hide in. But we don’t know the distribution (that is, which areas are high or low density).

## Balancing Privacy with Utility

We combined local noise addition with a technique called *secure aggregation* to address these concerns of balancing privacy with utility. To understand how balancing privacy with utility works, let’s dive into a Photos Memories use case. 

A user takes a photo in a place they visit. The photo is annotated with common categories, such as recreation, person, sky, and so on. The model that assigns those categories runs locally on the user’s device, as described in [A Multi-Task Neural Architecture for On-Device Scene Analysis.](https://machinelearning.apple.com/research/on-device-scene-analysis) 

If the user opts in to the Analytics & Improvements feature, and enables Location Services and precise location stored on photos, we randomly pick one location-category pair (like *Central Park (New York)* and *person*) and encode it into a one-hot vector, as described in [Figure 2](#figure2).

Now, we take this one-hot vector and flip each bit with some probability. The noise introduced by flipped bits provides a local DP assurance, which is later amplified through secure aggregation.

<Figure id="figure2" theme="dark-gray" className="full-bleed" title="Figure 2" caption="Figure 2: On the user’s device, one randomly selected location-category pair gets encoded into a one-hot vector. Each tile on the map shown here corresponds to a segment within the vector. Each segment contains one bit representing each of the categories. Then each of the bits is randomly flipped with some probability to satisfy the DP assurance.">
 <Figure2 title="Encoding Location-Category as One-Hot Vector" />
</Figure>

Next, we split this binary vector (one-hot vector with random noise) into two shares. Each share on its own is meaningless noise, but when combined with the other, we get the noised vector back. Each share is encrypted with a different public key and uploaded to the server.

<Figure id="figure3" className="inline" title="Figure 3" caption="Figure 3: The secure aggregation protocol that enforces the DP assurance. Each binary vector is split into two shares, and each share is encrypted with a different public key. As long as no single entity gets access to both private keys, nobody can see any individual vectors, only the aggregate, which satisfies the desired DP assurance.">
 <Image 
 src="https://mlr.cdn-apple.com/media/Fig_3_pipeline_38eaf7f2d0.png"
 />
</Figure>

On the server side, there are two main components, as shown in [Figure 3](#figure3): 

- **Leader** has one private key to decrypt, and it aggregates the first shares from the given cohort of devices. 
- **Helper** has the other private key to decrypt, and it aggregates the second shares from the given cohort of devices. 

Both the helper and leader components publish their aggregate only if it satisfies a minimal cohort size (the minimal number of devices contributing their shares). As long as no single entity has access to both private keys, nobody can see the original one-hot vector from any single device. They would see only the aggregates, which satisfies the DP assurance. This DP assurance is enforced through local noise (added on the user device), which gets amplified by secure aggregation, as described in [Private Federated Statistics in an Interactive Setting](https://arxiv.org/abs/2211.10082).

Finally, both aggregates from the helper and leader functions are combined to get a noisy aggregate vector from all the devices within the cohort. This vector is decoded back into the location and photo categories, which can then be visualized on a map, as shown in [Figure 4.](#figure4)

We needed a solution to address one more complication: Detecting malicious updates would be difficult because nobody can see any single vector. For example, a malicious user might submit vectors to the server that would poison the final histogram. We surmounted this obstacle using Prio validation, as discussed in the work [Prio: Private, Robust, and Scalable Computation of Aggregate Statistics.](https://arxiv.org/abs/1703.06255) 

<Figure id="figure4" className="" title="Figure 4" caption="Figure 4: Map showing final normalized histograms of top categories for photos taken near the Brooklyn Bridge in New York City.">
 <Image 
 src="https://mlr.cdn-apple.com/media/Fig_4_histograms_c0049fb8b0.jpeg" rounded
 darkgray
 />
</Figure>

# Providing Iconic Scenes Frequencies 

We discovered frequencies for 4.5 million location-category pairs for 1.5 million unique locations and 100 categories using this approach. These frequencies have been powering ML selections of Memories key photos for millions of device owners worldwide since iOS&nbsp;16, as well as a ranking of photos and locations in Places Map in iOS&nbsp;17.

The learned histograms satisfy DP $epsilon=1, delta=1.5e^{-7}$. We used a cohort size of 150,000. To avoid dropping histograms with smaller populations, we add additional noise in the secure aggregation protocol to achieve a DP of $epsilon=1$. To ensure our strong privacy assurance holds without requiring that users trust either the helper or the leader component, this noise is added independently by both the helper and leader components.

# Building on Our Wins and Looking Forward

Privacy-preserving ML research has already improved user experiences, and this is just the beginning. There are many opportunities to improve our system further, including:

- Privacy accounting and transparency
- Better algorithms
- Other data science tools

**Privacy accounting and transparency.** The current approach provides a formal DP assurance for a given histogram. However, DP is composable, and that’s an opportunity to provide an even more precise privacy assurance—one that is not specific to a single task or user data. One of the next steps is exploring formal accounting methods for precise accounting across multiple histograms. 

The current approach is compatible with the [Distributed Aggregation Protocol](https://datatracker.ietf.org/doc/draft-ietf-ppm-dap/) developed by the IETF working group, which is publicly available. We would like to open source key components of the system to make it easier, to verify the correctness of our implementation.

**Better algorithms.** The data is not uniformly distributed. For instance, many people might take pictures in one location, and very few might take photos in other locations. We are limited to fixed precision to provide a consistent privacy assurance with our current approach, which can’t be optimal for all locations. An iterative approach of gradually traversing the data and drilling into more popular places with higher precision will likely bring better results. 

Each device has hundreds or thousands of photos with multiple categories, but we pick only one random category from one randomly sampled historical photo. There is an opportunity to develop efficient algorithms for encoding and aggregating multiple location-category pairs to produce more precise histograms while maintaining the same privacy assurance.

**Other data science tools.** Histograms are very powerful, but to build even better experiences with privacy-preserving ML, we must provide a full set of data science tools. For example, for tasks that have good supervision, gradient-based algorithms might be powerful, along with tools for rigorous statistical testing to compare different solutions. And for tasks where no supervision is available, unsupervised algorithms such as k-means clustering might be needed. Many of those algorithms can be implemented using histograms, but custom solutions often bring significantly better utility, which is key for building better user experiences.

# Conclusion

Our goal is to provide an interpretable privacy assurance to our users, in a transparent way. As more users worldwide contribute their photos of iconic scenes in more locations, the resulting datasets will better represent the world’s population, enabling us to build more inclusive experiences.

In this post, we described a step toward that goal, how we learned frequencies of iconic scenes with formal DP assurance. This enabled us to improve key photo selection for Memories in iOS&nbsp;16, and Places in iOS&nbsp;17. This approach of applying privacy-preserving machine learning research to real-world problems fuels ML innovation and helps our users keep their data private.

# Acknowledgments

Many people contributed to this research, including Mike Chatzidakis, Junye Chen, Oliver Chick, Eric Circlaeays, Sowmya Gopalan, Yusuf Goren, Kristine Guo, Michael Hesse, Omid Javidbakht, Vojta Jina, Kalu Kalu, Anil Katti, Albert Liu, Richard Low, Audra McMillan, Joey Meyer, Steve Myers, Alex Palmer, David Park, Gianni Parsa, Paul Pelzl, Rehan Rishi, Michael Scaria, Chiraag Sumanth, Kunal Talwar, Karl Tarbe, Shan Wang, and Mayank Yadav.

# Apple Resources   

Apple. “Learning with Privacy at Scale." n.d. [[link.]](https://docs-assets.developer.apple.com/ml-research/papers/learning-with-privacy-at-scale.pdf)

Apple Support. “Use Memories in Photos on your iPhone, iPad, or iPod touch.” 2023. [[link.]](https://support.apple.com/en-us/HT207023)

McMillan, Audra, Omid Javidbakht, Kunal Talwar, Elliot Briggs, Mike Chatzidakis, Junye Chen, John Duchi, et al. 2022. “Private Federated Statistics in an Interactive Setting.” [[link.]](https://arxiv.org/abs/2211.10082)

“A Multi-Task Neural Architecture for On-Device Scene Analysis." 2022. Apple Machine Learning Research. [[link.]](https://machinelearning.apple.com/research/on-device-scene-analysis)

# External References  

Corrigan-Gibbs, Henry, and Dan Boneh. 2017. “Prio: Private, Robust, and Scalable Computation of Aggregate Statistics.” March 18, 2017. [[link.]](https://arxiv.org/abs/1703.06255)

Geoghegan, Tim, Christopher Patton, Eric Rescorla, and Christopher A. Wood. 2023. “Distributed Aggregation Protocol for Privacy Preserving Measurement.” IETF. July 10, 2023. [[link.]](https://datatracker.ietf.org/doc/draft-ietf-ppm-dap/)

In this article, we share how we apply differential privacy (DP) to learn about the kinds of photos people take at frequently visited locations (iconic scenes) without personally identifiable data leaving their device. This approach is used in several features in Photos, including choosing key photos for <a href="https://support.apple.com/en-us/HT207023" class="more">Memories,</a> and selecting key photos for locations in Places in iOS 17.

Learning Iconic Scenes with Differential Privacy 

<div class="callout">

## Schedule

### Sunday, December&nbsp;10
<ul class="links-stacked">
<li>EXPO TALK</li>
 <li><a href=https://nips.cc/Expo/Conferences/2023/talk%20panel/78247>On-device Personal Voice for Accessibility</a></li>
 <li>12:00 - 12:50 PM CST, Room 215-216</li>
<li>Jiangchuan Li</li>
</ul>

### Monday, December&nbsp;11
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://blackinai.github.io/#/workshop/bai2023>Black in AI</a></li>
 <li>8:15 AM - 3:30 PM CST, Room R02-R05</li>
</ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://www.latinxinai.org/neurips-2023>LatinX in AI</a></li>
 <li>8:15 AM - 3:30 PM CST, Room 217-219</li>
 <li>Miguel Angel Bautista Martin will be participating in the mentoring hour as a mentor.</li></ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://www.queerinai.com/neurips-2023>Queer in AI</a></li>
 <li>9:30 AM - 3:30 PM CST, Room R06-R09</li>
</ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://sites.google.com/umich.edu/wiml2023/>Women in Machine Learning</a></li>
 <li>8:00 AM - 3:30 PM CST, Great Hall</li>
 <li>Derek Smith will be hosting a roundtable on Internships at Apple.</li>
</ul>

### Tuesday, December&nbsp;12
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=https://neurips.cc/virtual/2023/poster/70343>Provable Incremental Learning in Transformers</a></li>
 <li>10:45 AM - 12:45 PM CST in Great Hall & Hall B1+B2 #818</li>
<li>Emmanuel Abbe, Samy Bengio, Enric Boix-Adsera, Etai Littwin, Joshua Susskind</li>
</ul>

<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=/research/agnostically-learning-single>Agnostically Learning Single-Index Models Using Omnipredictors</a></li>
 <li>10:45 AM - 12:45 CST in Great Hall & Hall B1+B2 #1800</li>
<li>Parikshit Goapalan, Aravind Gollakotta, Adam Klivans, Konstantinos Stavropoulos</li>
</ul>
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=/research/planner>PLANNER: Generating Diversified Paragraph via Latent Language Diffusion Model</a></li>
 <li>10:45 AM - 12:45 PM CST in Great Hall & Hall B1+B2 #1921</li>
<li>Yizhe Zhang, Jiatao Gu, Zhuofeng Wu, Shuangfei Zhai, Josh Susskind, Navdeep Jaitly</li>
</ul>

### Wednesday, December&nbsp;13
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=https://neurips.cc/virtual/2023/poster/71651>DeepPCR: Parallelizing Sequential Operations in Neural Networksy</a></li>
 <li>5:00 - 7:00 PM CST in Great Hall & Hall B1+B2 #1009</li>
<li>Federico Danieli, Miguel Sarabia del Castillo, Pau Rodriguez, <Xavier Suau Cuadros, Luca Zappella</li>
 <ul class="links-stacked">
<li>POSTER</li>
 <li><a href=https://neurips.cc/virtual/2023/poster/71544>4M: Massively Multimodal Masked Modeling</a></li>
 <li>5:00-7:00 PM CST in Great Hall & Hall B1+B2 #1022</li>
<li>David Mizrahi, Mingfei Gao, Afshin Dehghan, Roman Bachmann, Oğuzhan Fatih Kar, Teresa Yeo, Amir Zamir</li>
</ul>
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=https://neurips.cc/virtual/2023/poster/71052>Unbalanced Low-rank Optimal Transport Solvers</a></li>
 <li>5:00 - 7:00 PM CST in Great Hall & Hall B1+B2 #1012</li>
<li>Meyer Scetbon, Michal Klein, Giovanni Palla, Marco Cuturi</li>
</ul>
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=/research/adaptive-weight>Adaptive Weight Decay</a></li>
 <li>5:00 - 7:00 PM CST in Great Hall & Hall B1+B2 #721</li>
<li>Mohammad Amin Ghiasi, Ali Shafahim Reza Ardekani</li>
</ul>
</ul>

### Thursday, December&nbsp;14
<ul class="links-stacked">
<li>POSTER</li>
 <li><a href=/research/parameter-free>PDP: Parameter-free Differentiable Pruning is All You Need</a></li>
 <li>10:45 AM - 12:45 PM CST, Great Hall & Hall B1+B2 #516</li>
<li>Minsik Cho, Saurabh Adya, Devang Naik</li>
</ul>

### Friday, December&nbsp;15
<ul class="links-stacked">
 <li>WORKSHOP</li>
 <li><a href=https://ml4physicalsciences.github.io/2023/>Machine Learning and the Physical Sciences</a></li>
 <li>Tatiana Likhomanenko is a reviewer for this workshop.</li>

</ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://sites.google.com/view/distshift2023/schedule>DistShift</a></li>
 <li>9:00 AM - 5:00 PM CST, Room R06-R09</li>
<li>TiC-CLIP: Continual Training of CLIP Models</li>
<li>Saurabh Garg, Hadi Pour Ansari, Mehrdad Farajtabar, Sachin Mehta, Raviteja Vemulapalli, Oncel Tuzel, Vaishaal Shankar, Fartash Faghri</li>
</ul>

### Saturday, December&nbsp;16


<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://neurips.cc/virtual/2023/workshop/66505>Generative AI and Biology</a></li>
<li>Generating Molecular Conformers with Manifold Diffusion Fields</li>
<li>Yuyang Wang, Ahmed Elhag, Navdeep Jaitly, Josh Susskind, Miguel Angel Bautista Martin<li>
 <li>6:15 AM, Room 265 - 268</li> 
</ul>

<ul class="links-stacked">
<li>WORKSHOP</li>
<li><a href=https://sites.google.com/view/icbinb-2023/home>I Can’t Believe It’s Not Better (ICBINB): Failure Modes in the Age of Foundation Models</a></li>
<li>Pretrained Language Models Do Not Help Auto-regressive Text-to-Image Generation</li>
<li>Yuhui Zhang, Brandon McKinzie, Vaishaal Shankar, Zhe Gan, Alexander Toshev<li>
 <li>8:45 AM - 5:30 PM PT, Room R0T-R05</li>
</ul>

<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://neurips2023-enlsp.github.io>Efficient Natural Language and Speech Processing</a></li>
<li>8:30 AM - 5:10 CST, Room 206-207</li>
 <li><a href=/research/relu>ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models</a><li>Iman Mirzadeh, Keivan Alizadeh, Sachin Mehta, Carlo C Del Mundo, Oncel Tuzel, Golnoosh Samei, Mohammad Rastegari, Mehrdad Farajtabar</li>
<li>Structured Pruning Under Limited Data</li>
<li>Lucio Dery, Awni Hannun</li>
</ul>

<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://otmlworkshop.github.io/>Optimal Transport and Machine Learning</a></li>
 <li>Marco Cuturi is a co-organizer of this workshop.</li>
</ul>


<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://federated-learning.org/fl@fm-neurips-2023>International Workshop on Federated Learning in the Age of Foundation Models</a></li>
 <li>4:15-5:30 PM CS in Hall D-2</li>
 <li>Federated Learning with Differential Privacy for ASR</li>
<li>Martin Pelikan, Shams Azam, Honza Silovsky, Kunal Talwar, Vitaly Feldman, Tatiana Likhomanenko</li>
 <li>Privately Estimating Personalized Histograms in a Federated Setting</li>
 <li>Amrith Setlur, Vitaly Feldman, Kunal Talwar</li>
</ul>

<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://sslneurips23.github.io/pages/call-for-participation.html>Self-Supervised Learning Theory and Practice</a></li>
<li>Bootstrap Your Own Variance</li>
<li>Polina Turishcheva, Jason Ramapuram, Sinead Williamson, Dan Busbridge, Eeshan Gunesh Dhekane, Russ Webb<li>
 <li>12:00 - 1:00 PM CST and 4:15 - 5:30 PM CST, Room 217-219</li>
</ul>

</div>

## Accepted Papers

[4M: Massively Multimodal Masked Modeling](/research/massively-multimodal)

*David Mizrahi, Mingfei Gao, Afshin Dehghan, Roman Bachmann, Oğuzhan Fatih Kar, Teresa Yeo, Amir Zamir*

[Adaptive Weight Decay](/research/adaptive-weight)

*Amin Ghiasi, Ali Shafahi, Reza Ardekani*

[Agnostically Learning Single-Index Models using Omnipredictors](/research/agnostically-learning-single)

*Aravind Gollakota, Parikshit Gopalan, Adam R. Klivans, Konstantinos Stavropoulos*

[PLANNER: Generating Diversified Paragraph via Latent Language Diffusion Model](/research/planner)

*Yizhe Zhang, Jiatao Gu, Zhuofeng Wu, Shuangfei Zhai, Josh Susskind, Navdeep Jaitly*

[PDP: Parameter-free Differentiable Pruning is All You Need](/research/parameter-free)

*Minsik Cho, Saurabh Adya, Devang Naik*

[Swap Agnostic Learning, or Characterizing Omniprediction via Multicalibration](/research/swap-agnostic-learning)

*Parikshit Gopalan, Michael P. Kim, Omer Reingold*

## Workshop Accepted Papers

[ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Model](/research/relu)

*Iman Mirzadeh, Keivan Alizadeh, Sachin Mehta, Carlo C Del Mundo, Oncel Tuzel, Golnoosh Samei, Mohammad Rastegari, Mehrdad Farajtabar*

[SAM-CLIP: Merging Vision Foundation Models towards Semantic and Spatial Understanding](/research/sam-clip)

*Haoxiang Wang, Pavan Kumar Anasosalu Vasu, Fartash Faghri, Raviteja Vemulapalli, Mehrdad Farajtabar, Sachin Mehta, Mohammad Rastegari, Oncel Tuzel, Hadi Pouransari*

[What Algorithms can Transformers Learn? A Study in Length Generalization](/research/transformers-learn)

*Hattie Zhou, Omid Saremi, Etai Littwin, Arwen Bradley, Noam Razin, Josh Susskind, Samy Bengio, Preetum Nakkiran*

## Demos

On-device Personal Voice for Accessibility

On-device ML model training is a key research area we focus on. In this demo we will share how we applied text-to-speech model adaptation technology to Apple devices to build the personal voice with limited number of recordings, so that the people who are at risk of losing their voice can store their voice and use it in live speech when they are not able to speak.

Back to Back Siri Requests

Siri Directed Speech Detector (SDSD) runs completely on device. It makes use of information from acoustics, automatic speech recognition system, and lexical information to decide if an utterance is directed towards Siri or not. With this feature, Siri listens to user speech for a subsequent request, while it is responding to the previous request.

All NeurIPS attendees are invited to stop by the Apple booth (booth number 1103, located in Hall C and D of the New Orleans Ernest N. Morial Convention Center) to experience these demos in person.

## Acknowledgements

Samy Bengio is a member of the NeurIPS 2023 Board.

Marian Stewart Bartlett is the NeurIPS 2023 Treasurer.

Alexander Toshev, Kunal Talwar, Navdeep Jaitly, Pierre Ablin, Preetum Nakkiran, Shuangfei Zhai, Vaishaal Shankar, Vitaly Feldman, and Yizhe Zhang are Area Chairs for NeurIPS 2023.

Marco Cuturi and Samy Bengio are NeurIPS 2023 Senior Area Chairs.

Audra McMillan is an Ethics Reviewer for NeurIPS 2023.

Tatiana Likhomanenko and Oncel Tuzel are NeurIPS 2023 Datasets and Benchmarks Area Chairs.

Aravind Gollakota, Arno Blaas, Barry Theobald, Bhavika Devnani, Bogdan Mazoure, Enrico Fini, Etai Littwin, Hilal Asi, Jason Ramapuram, Lyndon Duong, Miguel Sarabia, Pau Rodriguez, Rahul Sunil Bhalerao, Richard Bai, Skyler Seto, Tatiana Likhomanenko, Vimal Thilak, and Xavi Suau are Reviewers for NeurIPS 2023.

Apple is sponsoring the Neural Information Processing Systems (NeurIPS) conference, which will take place in person from December 10 to 16 in New Orleans, Louisiana. 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. Below is the schedule of Apple-sponsored workshops and events at NeurIPS 2023.

Neural Information Processing Systems (NeurIPS) 2023 

<div class="callout">

## Schedule

### Wednesday, December&nbsp;6
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://www.winlp.org/winlp-2023-workshop/>Widening Natural Language Processing</a></li>
 <li>8:30 AM - 5:00 PM LT</li>

<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://sites.google.com/view/crac2023/>Computational Models of Reference, Anaphora and Coreference</a></li>
 <li>2:00-2:10 PM</li>
 <li><a href=/research/marrs>Accepted Paper: MARRS: Multi-Modal Reference Resolution Service</a></li>
</ul>
</ul>

### Thursday, December&nbsp;7
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://blackboxnlp.github.io>BlackboxNLP</a></li>
<li>9:00 AM - 6:00 PM LT, West Ballroom 2</li>
</ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://code-switching.github.io/2023>Computational Approaches to Linguistic Code-Switching</a></li>
<li>9:00 AM - 12:30 PM LT, West Ballroom 3</li>
<li><a href=https://machinelearning.apple.com/research/towards-real-world>Accepted Paper: Towards Real-World Streaming Speech Translation for Code-Switched Speech</a></li>
</ul>
<ul class="links-stacked">
<li>WORKSHOP</li>
 <li><a href=https://sites.google.com/view/crac2023/>Computational Models of Reference, Anaphora and Coreference</a></li>
<li>9:00 AM - 12:30 PM LT, West Ballroom 3</li>
 <li><a href=/research/marrs>Accepted Paper: MARRS: Multi-Modal Reference Resolution Service</a></li>
</ul>

### Friday, December&nbsp;8
<ul class="links-stacked">
<li>ORAL</li>
 <li><a href=/research/fleek>FLEEK: Factual Error Detection and Correction with Evidence Retrieved from External Knowledge</a></li>
 <li>11:00 AM SLT, East Foyer at Resorts World Convention Centre</li>
 <li>Kun Qian</li>
</ul>

### Sunday, December&nbsp;10
<ul class="links-stacked">
<li>ORAL</li>
 <li><a href=/research/steer>STEER: Semantic Turn Extension-Expansion Recognition for Voice Assistants</a></li>
 <li>11:00 AM - 12:30 PM LT, East Foyer, Aper ID Industry 170</li>
 <li>Leon Liyang Zhang, Jiarui Lu, Joel Ruben Antony Moniz, Aditya Kulkarni, Dhivya Piraviperumal, Tien Dung Tran, Nicholas Tzou, Hong Yu</li>
<li>ORAL</li>
 <li><a href=/research/eelbert>EELBERT: Tiny Models through Dynamic Embeddings</a></li>
 <li>11:00 AM - 12:30 PM LT, East Foyer, Paper ID Industry 129</li>
 <li>Gabrielle Cohn, Rishika Agarwal, Deepanshu Gupta, Siddharth Patwardhan</li>
<li>ORAL</li>
 <li><a href=/research/delphi>DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues</a></li>
 <li>11:00 AM - 12:30 PM LT, East Foyer, Paper ID Industry 207</li>
 <li>David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein</li>
</ul>
</div> 

## Accepted Papers

[DELPHI: Data for Evaluating LLMs' Performance in Handling Controversial Issues](/research/delphi)

*David Q. Sun, Artem Abzaliev, Hadas Kotek, Zidi Xiu, Christopher Klein*

[EELBERT: Tiny Models through Dynamic Embeddings](/research/eelbert)

*Gabrielle Cohn, Rishika Agarwal, Deepanshu Gupta, Siddharth Patwardhan*

[Increasing Coverage and Precision of Multilingual Textual Information in Knowledge Graphs](/research/increasing-coverage)

*Simone Conia, Min Li, Daniel Lee, Umar Farooq Minhas, Ihab Ilyas, Yunyao Li*

[FLEEK: FactuaL Error Detection and Correction with Evidence Retrieved from External Knowledge](/research/fleek)

*Farima Fatahi Bayat, Kun Qian, Benjamin Han, Yisi Sang, Anton Belyi, Samira Khorshidi, Fei Wu, Ihab Ilyas, Yunyao Li*

[STEER: Semantic Turn Extension-Expansion Recognition for Voice Assistants](/research/steer)

*Leon Liyang Zhang, Jiarui Lu, Joel Ruben Antony Moniz, Aditya Kulkarni, Dhivya Piraviperumal, Tien Dung Tran, Nicholas Tzou, Hong Yu*

## Workshop Accepted Papers

[MARRS: Multi-Modal Reference Resolution Service](/research/marrs)

*Anand Dhoot, Andy Tseng, Ankit Samal, Dhivya Piraviperumal, Halim Cagri Ates, Hong Yu, Jiarui Lu, Joel Moniz, Melis Ozyildirim, Roman Nguyen, Shirley (Rong) Zou, Shruti Bhargava, Siddardha Maddula, Site Li, Thy Tran, Yuan Zhang*

[Towards Real-World Streaming Speech Translation for Code-Switched Speech](/research/towards-real-world)

*Belen Alastruey, Matthias Sperber, Christian Gollan, Dominic Telaar, Tim Ng, Aashish Agargwal*

## Acknowledgements

Reza Shirani, Sid Patwardhan and Yizhe Zhangare reviewers for EMNLP 2023.

Apple is sponsoring the Empirical Methods in Natural Language Processing (EMNLP) conference, which will take place in person from December 6 to 10 in Singapore. EMNLP is a leading conference focused on natural language processing. Below is the schedule of Apple-sponsored workshops and events at EMNLP 2023.

Empirical Methods in Natural Language Processing (EMNLP) 2023

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

Overview

This paper was accepted at the NeurIPS 2023 workshop on Diffusion Models. 

We demonstrate how conditional generation from diffusion models can be used to tackle a variety of realistic tasks in the production of music in 44.1kHz stereo audio with sampling-time guidance. The scenarios we consider include continuation, inpainting and regeneration of musical audio, the creation of smooth transitions between two different music tracks, and the transfer of desired stylistic characteristics to existing audio clips. We achieve this by applying guidance at sampling time in a simple framework that supports both reconstruction and classification losses, or any combination of the two. This approach ensures that generated audio can match its surrounding context, or conform to a class distribution or latent representation specified relative to any suitable pre-trained classifier or embedding model. 

We show randomly chosen samples for a number of creative applications in Table 1, each conditioned on a given audio prompt. For each task and prompt we show samples from the different models described in the paper.

Task types:
- infill: replace the middle two seconds of the prompt
- regeneration: regenerate the middle two seconds of the prompt
- continuation: generate a new continuation starting from the first 2.4s of the prompt
- transitions: regenerate a crossfaded section between two tracks
- guidance: generate a new clip conditioned on the [PaSST](https://github.com/kkoutini/PaSST) classifier embedding of the prompt


<Figure className="inline" title="Table 1" id="table1" caption={<>Table 1: randomly selected samples for each task and prompt.</>}>
<div className="table-wrapper minimal">
<table aria-label="Table 1: randomly selected samples for each task and prompt.">
 <thead>
 <tr>
 <th>prompt</th>
 <th>task</th>
 <th>CQTDiff (baseline)</th>
 <th>latent</th>
 <th>waveform</th>
 </tr>
</thead>
<tbody>
 <tr>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/infill_original_none_750_8ac92c9d11.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td class="align-middle">infill</td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/A2_infill_cqt_diff_ddpm50_d6yt8gkhaj_750_80241756f5.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/A3_infill_latent_fma_ddpm500_w6fbs3urfc_750_1449e24420.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/A4_infill_waveform_fma_d12_ddpm500_huzh4umzgx_750_314ef9b2af.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/B1_infill_original_none_871_3005a7e59f.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>infill</td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/B2_infill_cqt_diff_ddpm50_d6yt8gkhaj_871_104839b88f.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/B3_infill_latent_fma_ddpm500_w6fbs3urfc_871_6d9bc39473.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/B4_infill_waveform_fma_d12_ddpm500_huzh4umzgx_871_f88f1d9373.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/C1_infill_original_none_956_d8f14ecacd.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>infill</td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/C2_infill_cqt_diff_ddpm50_d6yt8gkhaj_956_5f594a4a13.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/C3_infill_latent_fma_ddpm500_w6fbs3urfc_956_ada6ff467e.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/C4_infill_waveform_fma_d12_ddpm500_huzh4umzgx_956_a4967d25e1.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/D1_regenerate_original_none_819_fdd7e0eb06.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>regenerate</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/D3_regenerate_latent_fma_ddpm50_cd64hrxnzu_819_2b7f94e6b9.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/D4_regenerate_waveform_fma_d12_ddpm50_mef3wpz8b4_819_8a092c2b55.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/E1_regenerate_original_none_738_fe2ea3982c.mp3" type="audio/mpeg"/>
 </audio>
 </td>
 <td>regenerate</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/E3_regenerate_latent_fma_ddpm50_cd64hrxnzu_738_379d60d04a.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/E4_regenerate_waveform_fma_d12_ddpm50_mef3wpz8b4_738_98a2d7a9f0.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/F1_regenerate_original_none_809_5a244adb31.mp3" type="audio/mpeg"/>
 </audio>
 </td>
 <td>regenerate</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/F3_regenerate_latent_fma_ddpm50_cd64hrxnzu_809_d003c3ac98.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/F4_regenerate_waveform_fma_d12_ddpm50_mef3wpz8b4_809_498dafcb0b.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/G1_continuation_original_none_871_763713c11c.mp3" type="audio/mpeg"/>
 </audio>
 </td>
 <td>continuation</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/G3_continuation_latent_fma_ddpm500_mrwwprarm4_871_5a5921ac72.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/G4_continuation_waveform_fma_d12_ddpm500_7zapupbgds_871_13b5da6562.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/H1_continuation_original_none_790_a5572f81a0.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>continuation</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/H3_continuation_latent_fma_ddpm500_mrwwprarm4_790_9c52f78ac8.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/H4_continuation_waveform_fma_d12_ddpm500_7zapupbgds_790_f0c5dfcb96.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/I1_continuation_latent_fma_ddpm500_mrwwprarm4_970_b86e55f604.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>continuation</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/I3_continuation_latent_fma_ddpm500_mrwwprarm4_970_0d0ab80ca1.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/I4_continuation_waveform_fma_d12_ddpm500_7zapupbgds_970_8be16db1fe.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/J1_transitions_original_none_121_7f328cdf89.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>transitions</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/J3_transitions_latent_fma_ddpm50_438ns6pvwt_121_dcfb73dc5c.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/J4_transitions_waveform_fma_d12_ddpm50_54tm5d9qvn_121_c22d714fcb.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/K1_transitions_original_none_144_8ba9792d75.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>transitions</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/K3_transitions_latent_fma_ddpm50_438ns6pvwt_144_d0d25d66e2.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/K4_transitions_waveform_fma_d12_ddpm50_54tm5d9qvn_144_caebb426b0.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/L1_transitions_original_none_91_fe68076b2b.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>transitions</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/L3_transitions_latent_fma_ddpm50_438ns6pvwt_91_99da2c4b63.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/L4_transitions_waveform_fma_d12_ddpm50_54tm5d9qvn_91_01b6f18107.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/M1_guidance_original_none_790_7ee94d4443.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>guidance</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/M3_guidance_latent_fma_ddpm500_3ivp9uccu5_790_48cc1782f2.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/M4_guidance_waveform_fma_d12_ddpm500_kqigx8mpxh_790_959fca056d"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/N1_guidance_original_none_925_3d3378d4b8.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>guidance</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/N3_guidance_latent_fma_ddpm500_3ivp9uccu5_925_ade1c6b161.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/N4_guidance_waveform_fma_d12_ddpm500_kqigx8mpxh_925_a06f199434.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 <tr>
 <td>
 <audio controls>
 <source src="https://mlr.cdn-apple.com/media/O1_guidance_original_none_29_cf38eea302.mp3" type="audio/mpeg" />
 </audio>
 </td>
 <td>guidance</td>
 <td></td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/O3_guidance_latent_fma_ddpm500_3ivp9uccu5_29_df9b2da8f7.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 <td>
 <audio controls>
 <source
 src="https://mlr.cdn-apple.com/media/O4_guidance_waveform_fma_d12_ddpm500_kqigx8mpxh_29_76631c6c10.mp3"
 type="audio/mpeg"
 />
 </audio>
 </td>
 </tr>
 </tbody>
</table>
</div></Figure>

Prompts are drawn from a test split of the [Free Music Archive dataset](https://github.com/mdeff/fma), published by Michaël Defferrard et al. under a [Creative Commons Attribution 4.0 International License (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0).

NeurIPS

Controllable Music Production with Diffusion Models and Guidance Gradients

This paper was accepted at Generative AI and Biology workshop at NeurIPS 2023. 

In this paper we tackle the problem of generating a molecule conformation in 3D space given its 2D structure. We approach this problem through the lens of a diffusion model for functions in Riemannian Manifolds. Our approach is simple and scalable, and obtains results that are on par with state-of-the-art while making no assumptions about the explicit structure of molecules. 

Machine Learning
Research at Apple

Recent research

Controllable Music Production with Diffusion Models and Guidance Gradients

Generating Molecular Conformers with Manifold Diffusion Fields

Research highlights

Voice Trigger System for Siri

Learning Iconic Scenes with Differential Privacy

All events

Neural Information Processing Systems (NeurIPS) 2023

Empirical Methods in Natural Language Processing (EMNLP) 2023

Discover opportunities in Machine Learning.

Machine LearningResearch at Apple

Recent research

Controllable Music Production with Diffusion Models and Guidance Gradients

Generating Molecular Conformers with Manifold Diffusion Fields

Research highlights

Voice Trigger System for Siri

Learning Iconic Scenes with Differential Privacy

All events

Neural Information Processing Systems (NeurIPS) 2023

Empirical Methods in Natural Language Processing (EMNLP) 2023

Discover opportunities in Machine Learning.

Machine Learning
Research at Apple