Recommendation systems in large-scale online marketplaces are essential to aiding users in discovering new content. However, state-of-the-art systems for item-to-item recommendation tasks are often based on a shallow level of contextual relevance, which can make the system insufficient for tasks where item relationships are more nuanced. Contextually relevant item pairs can sometimes have problematic relationships that are confusing or even controversial to end users, and they could degrade user experiences and brand perception when recommended to users. For example, the recommendation of a book about one sports team to someone reading a book about that team’s biggest rival could be a bad experience, despite the presumed similarities of the books. In this paper, we propose a classifier to identify and prevent such problematic item-to-item recommendations and to enhance overall user experiences. The proposed approach utilizes active learning to sample hard examples effectively across sensitive item categories and employs human raters for data labeling. We also perform offline experiments to demonstrate the efficacy of this system for identifying and filtering problematic recommendations while maintaining recommendation quality.

In this work, we provide an in-depth characterization study of the performance overhead for running Transformer models with secure multi-party computation (MPC). MPC is a cryptographic framework for protecting both the model and input data privacy in the presence of untrusted compute nodes. Our characterization study shows that Transformers introduce several performance challenges for MPC-based private machine learning inference. First, Transformers rely extensively on “softmax” functions. While softmax functions are relatively cheap in a non-private execution, softmax dominates the MPC inference runtime, consuming up to 50% of the total inference runtime. Further investigation shows that computing the maximum, needed for providing numerical stability to softmax, is a key culprit for the increase in latency. Second, MPC relies on approximating non-linear functions that are part of the softmax computations, and the narrow dynamic ranges make optimizing softmax while maintaining accuracy quite difficult. Finally, unlike CNNs, Transformer-based NLP models use large embedding tables to convert input words into embedding vectors. Accesses to these embedding tables can disclose inputs; hence, additional obfuscation for embedding access patterns is required for guaranteeing the input privacy. One approach to hide address accesses is to convert an embedding table lookup into a matrix multiplication. However, this naive approach increases MPC inference runtime significantly. We then apply tensor-train (TT) decomposition, a lossy compression technique for representing embedding tables, and evaluate its performance on embedding lookups. We show the trade-off between performance improvements and the corresponding impact on model accuracy using detailed experiments.

Secure multi-party computation (MPC) allows parties to perform computations on data while keeping that data private. This capability has great potential for machine-learning applications: it facilitates training of machine-learning models on private data sets owned by different parties, evaluation of one party's private model using another party's private data, etc. Although a range of studies implement machine-learning models via secure MPC, such implementations are not yet mainstream. Adoption of secure MPC is hampered by the absence of flexible software frameworks that `"speak the language" of machine-learning researchers and engineers. To foster adoption of secure MPC in machine learning, we present CrypTen: a software framework that exposes popular secure MPC primitives via abstractions that are common in modern machine-learning frameworks, such as tensor computations, automatic differentiation, and modular neural networks. This paper describes the design of CrypTen and measure its performance on state-of-the-art models for text classification, speech recognition, and image classification. Our benchmarks show that CrypTen's GPU support and high-performance communication between (an arbitrary number of) parties allows it to perform efficient private evaluation of modern machine-learning models under a semi-honest threat model. For example, two parties using CrypTen can securely predict phonemes in speech recordings using Wav2Letter faster than real-time. We hope that CrypTen will spur adoption of secure MPC in the machine-learning community.

We introduce CryptGPU, a system for privacy-preserving machine learning that implements all operations on the GPU (graphics processing unit). Just as GPUs played a pivotal role in the success of modern deep learning, they are also essential for realizing scalable privacy-preserving deep learning. In this work, we start by introducing a new interface to losslessly embed cryptographic operations over secret-shared values (in a discrete domain) into floating-point operations that can be processed by highly-optimized CUDA kernels for linear algebra. We then identify a sequence of "GPU-friendly" cryptographic protocols to enable privacy-preserving evaluation of both linear and non-linear operations on the GPU. Our microbenchmarks indicate that our private GPU-based convolution protocol is over 150× faster than the analogous CPU-based protocol; for non-linear operations like the ReLU activation function, our GPU-based protocol is around 10× faster than its CPU analog. With CryptGPU, we support private inference and training on convolutional neural networks with over 60 million parameters as well as handle large datasets like ImageNet. Compared to the previous state-of-the-art, our protocols achieve a 2× to 8× improvement in private inference for large networks and datasets. For private training, we achieve a 6× to 36× improvement over prior state-of-the-art. Our work not only showcases the viability of performing secure multiparty computation (MPC) entirely on the GPU to newly enable fast privacy-preserving machine learning, but also highlights the importance of designing new MPC primitives that can take full advantage of the GPU’s computing capabilities.

In our daily lives, we rely heavily on our visual and auditory channels to receive information from others. In the case of impairment, or when large amounts of information are already transmitted visually or aurally, alternative methods of communication are needed. A haptic language offers the potential to provide information to a user when visual and auditory channels are unavailable. Previously created haptic languages include deconstructing acoustic signals into features and displaying them through a haptic device, and haptic adaptations of Braille or Morse code; however, these approaches are unintuitive, slow at presenting language, or require a large surface area. We propose using a multi-sensory haptic device called MISSIVE, which can be worn on the upper arm and is capable of producing brief cues, sufficient in quantity to encode the full English phoneme set. We evaluated our approach by teaching subjects a subset of 23 phonemes, and demonstrated an 86% accuracy in a 50 word identification task after 100 minutes of training.

Spherical data compression methods for Virtual Reality (VR) currently leverage popular rectangular data encoding algorithms. Traditional compression algorithms have massive adoption and hardware support on computers and mobile devices. Efficiently utilizing these two-dimensional compression methods for spherical data necessitates a projection from the three-dimensional surface of a sphere to a two-dimensional rectangle. Any such projection affects the final resolution distribution of the data after decoding. Popular projections used for VR video benefit from mathematical or geometric simplicity, but result in suboptimal resolution distributions. We introduce a method for generating a projection to match a desired resolution function. This method allows for customized projections with smooth, continuous and optimal resolution functions. Compared to commonly used projections, our resolution-defined projections drastically improve compression ratios for any given quality.

We consider one single antenna transmitter communicating with a receiver with a large number of antennas. Motivated by the optimal noncoherent detector in a Rayleigh fading channel, we propose a noncoherent energy-based communication scheme that does not require knowledge of instantaneous CSI (channel state information) at either the transmitter or the receiver; it uses only the statistics of the channel and noise. We explore the impact of coding to reduce the number of antennas needed for this system to achieve a given performance target. In particular, random coding error exponents for this system are used to determine tradeoff curves between the number of antennas and the blocklengths associated with a guaranteed performance target. However, since random codes have exponentially increasing decoding complexity with increasing blocklength, we also consider a simplified codebook design that has significantly lower encoding and decoding complexity. Simulations suggest that for small blocklengths, the performance of this simplified codebook is competitive with random coding constructions.