Machine Learning Research at Apple

MAY 1, 2024

Large Language Models (LLMs) with billions of parameters have drastically transformed AI applications. However, their demanding computation during inference has raised significant challenges for deployment on resource-constrained devices. Despite recent trends favoring alternative activation functions such as GELU or SiLU, known for increased computation, this study strongly advocates for reinstating ReLU activation in LLMs.

AI 147

AI AI 147

Machine Learning Research at Apple

APRIL 23, 2024

Understanding human motion from video is crucial for applications such as pose estimation, mesh recovery, and action recognition. While state-of-the-art methods predominantly rely on Transformer-based architectures, these approaches have limitations in practical scenarios. They are notably slower when processing a continuous stream of video frames in real time and do not adapt to new frame rates.

130

130

Machine Learning Research at Apple

APRIL 30, 2024

Contrastive learning typically matches pairs of related views among a number of unrelated negative views. Views can be generated (e.g. by augmentations) or be observed. We investigate matching when there are more than two related views which we call poly-view tasks, and derive new representation learning objectives using information maximization and sufficient statistics.

147

147

Machine Learning Research at Apple

APRIL 30, 2024

We investigate the capabilities of transformer models on relational reasoning tasks. In these tasks, models are trained on a set of strings encoding abstract relations, and are then tested out-of-distribution on data that contains symbols that did not appear in the training dataset. We prove that for any relational reasoning task in a large family of tasks, transformers learn the abstract relations and generalize to the test set when trained by gradient descent on sufficiently large quantities o

147

147

Machine Learning Research at Apple

APRIL 28, 2024

Sleep staging is a clinically important task for diagnosing various sleep disorders but remains challenging to deploy at scale because it requires clinical expertise, among other reasons. Deep learning models can perform the task but at the expense of large labeled datasets, which are unfeasible to procure at scale. While self-supervised learning (SSL) can mitigate this need, recent studies on SSL for sleep staging have shown performance gains saturate after training with labeled data from only

Supervised Learning 147

Supervised Learning Deep Learning Deep Learning 147

Machine Learning Research at Apple

APRIL 14, 2024

Pretrained language models are commonly adapted to comply with human intent and downstream tasks via finetuning. The finetuning process involves supervised finetuning (SFT), using labeled samples, and/or reinforcement learning based fine-tuning (RFT) via policy gradient methods, using a (possibly learned) reward function. This work highlights an overlooked optimization hurdle in RFT: we prove that the expected gradient for an input sample (i.e. prompt) vanishes if its reward standard deviation u

130

130

Machine Learning Research at Apple

APRIL 23, 2024

Neural knowledge-to-text generation models often struggle to faithfully generate descriptions for the input facts: they may produce hallucinations that contradict the given facts, or describe facts not present in the input. To reduce hallucinations, we propose a novel decoding method, TWEAK (Think While Effectively Articulating Knowledge). TWEAK treats the generated sequences at each decoding step and its future sequences as hypotheses, and ranks each generation candidate based on how well their

147

147

Machine Learning Research at Apple

APRIL 24, 2024

Contrastive learning has emerged as a transformative method for learning effective visual representations through the alignment of image and text embeddings. However, pairwise similarity computation in contrastive loss between image and text pairs poses computational challenges. This paper presents a novel weakly supervised pre-training of vision models on web-scale image-text data.

147

147

Machine Learning Research at Apple

APRIL 7, 2024

On-device machine learning (ML) promises to improve the privacy, responsiveness, and proliferation of new, intelligent user experiences by moving ML computation onto everyday personal devices. However, today's large ML models must be drastically compressed to run efficiently on-device, a hurtle that requires deep, yet currently niche expertise. To engage the broader human-centered ML community in on-device ML experiences, we present the results from an interview study with 30 experts at Apple th

Machine Learning 130

Machine Learning Machine Learning ML ML 130

Machine Learning Research at Apple

APRIL 23, 2024

On-device machine learning (ML) moves computation from the cloud to personal devices, protecting user privacy and enabling intelligent user experiences. However, fitting models on devices with limited resources presents a major technical challenge: practitioners need to optimize models and balance hardware metrics such as model size, latency, and power.

Machine Learning 147

Machine Learning Machine Learning ML ML 147

Machine Learning Research at Apple

APRIL 23, 2024

The reproducibility and transparency of large language models are crucial for advancing open research, ensuring the trustworthiness of results, and enabling investigations into data and model biases, as well as potential risks. To this end, we release OpenELM, a state-of-the-art open language model. OpenELM uses a layer-wise scaling strategy to efficiently allocate parameters within each layer of the transformer model, leading to enhanced accuracy.

354

354

Machine Learning Research at Apple

APRIL 14, 2024

Inspired by the advancements in foundation models for language-vision modeling, we explore the utilization of transformers and large-scale pretraining on biosignals. In this study, our aim is to design a general-purpose architecture for biosignals that can be easily trained on multiple modalities and can be adapted to new modalities or tasks with ease.

130

130

Machine Learning Research at Apple

APRIL 21, 2024

Adaptive gradient methods, notably Adam, have become indispensable for optimizing neural networks, particularly in conjunction with Transformers. In this paper, we present a novel optimization anomaly called the Slingshot Effect, which manifests during extremely late stages of training. We identify a distinctive characteristic of this phenomenon through cyclic phase transitions between stable and unstable training regimes, as evidenced by the cyclic behavior of the norm of the last layer’s weigh

147

147

Machine Learning Research at Apple

APRIL 14, 2024

Existing vision-language models exhibit strong generalization on a variety of visual domains and tasks. However, such models mainly perform zero-shot recognition in a closed-set manner, and thus struggle to handle open-domain visual concepts by design. There are recent finetuning methods, such as prompt learning, that not only study the discrimination between in-distribution (ID) and out-of-distribution (OOD) samples, but also show some improvements in both ID and OOD accuracies.

165

165

Machine Learning Research at Apple

APRIL 14, 2024

Long prompts present a significant challenge for practical LLM-based systems that need to operate with low latency and limited resources. We investigate prompt compression for zero-shot dialogue systems that learn to use unseen APIs directly in-context from their documentation, which may take up hundreds of prompt tokens per API. We start from a recently introduced approach (Mu et al., 2023) that learns to compress the prompt into a few “gist token” activations during finetuning.

147

147

Machine Learning Research at Apple

APRIL 15, 2024

Preference based Reinforcement Learning (PbRL) has shown great promise in learning from human preference binary feedback on agent's trajectory behaviors, where one of the major goals is to reduce the number of queried human feedback. While the binary labels are a direct comment on the goodness of a trajectory behavior, there is still a need for resolving credit assignment especially in limited feedback.

147

147

Machine Learning Research at Apple

MARCH 20, 2024

Keeping large foundation models up to date on latest data is inherently expensive. To avoid the prohibitive costs of constantly retraining, it is imperative to continually train these models. This problem is exacerbated by the lack of any large scale continual learning benchmarks or baselines. We introduce the first set of web-scale Time-Continual (TiC) benchmarks for training vision-language models: TiC-DataComp, TiC-YFCC, and TiC-Redcaps.

130

130

Machine Learning Research at Apple

APRIL 3, 2024

*Equal Contributors Contrastive pretraining of image-text foundation models, such as CLIP, demonstrated excellent zero-shot performance and improved robustness on a wide range of downstream tasks. However, these models utilize large transformer-based encoders with significant memory and latency overhead which pose challenges for deployment on mobile devices.

162

162

Machine Learning Research at Apple

MARCH 28, 2024

Neural Network Language Models (NNLMs) of Virtual Assistants (VAs) are generally language-, region-, and in some cases, device-dependent, which increases the effort to scale and maintain them. Combining NNLMs for one or more of the categories could be one way to improve scalability. In this work, we combine regional variants of English by building a "World English" NNLM.

147

147

Machine Learning Research at Apple

MARCH 31, 2024

Streaming neural network models for fast frame-wise responses to various speech and sensory signals are widely adopted on resource-constrained platforms. Hence, increasing the learning capacity of such streaming models (i.e., by adding more parameters) to improve the predictive power may not be viable for real-world tasks. In this work, we propose a new loss, Streaming Anchor Loss (SAL), to better utilize the given learning capacity by encouraging the model to learn more from essential frames.

147

147

Machine Learning Research at Apple

MARCH 21, 2024

We present an architecture for device-directed speech detection that treats the task as a text-generation problem. We use a multi-modal fusion approach that combines acoustic information from the recorded audio waveform with text and confidence information obtained from an automatic speech recognition system. The audio waveform is represented as a sequence of continuous embeddings by an audio encoder and presented as a prefix token to a pretrained large language model (LLM).

162

162

Machine Learning Research at Apple

MARCH 19, 2024

In this work, we discuss building performant Multimodal Large Language Models (MLLMs). In particular, we study the importance of various architecture components and data choices. Through careful and comprehensive ablations of the image encoder, the vision language connector, and various pre-training data choices, we identified several crucial design lessons.

162

162

Machine Learning Research at Apple

MARCH 14, 2024

Datasets that pair Knowledge Graphs (KG) and text together (KG-T) can be used to train forward and reverse neural models that generate text from KG and vice versa. However models trained on datasets where KG and text pairs are not equivalent can suffer from more hallucination and poorer recall. In this paper, we verify this empirically by generating datasets with different levels of noise and find that noisier datasets do indeed lead to more hallucination.

147

147

Machine Learning Research at Apple

MARCH 13, 2024

Wearable sensors have permeated into people's lives, ushering impactful applications in interactive systems and activity recognition. However, practitioners face significant obstacles when dealing with sensing heterogeneities, requiring custom models for different platforms. In this paper, we conduct a comprehensive evaluation of the generalizability of motion models across sensor locations.

147

147

Machine Learning Research at Apple

MARCH 13, 2024

While Automatic Speech Recognition (ASR) systems are widely used in many real-world applications, they often do not generalize well to new domains and need to be finetuned on data from these domains. However, target-domain data is usually not readily available in many scenarios. In this paper, we propose a new strategy for adapting ASR models to new target domains without any text or speech from those domains.

147

147

Machine Learning Research at Apple

MARCH 3, 2024

Human following serves an important human-robotics interaction feature, while real-world scenarios make it challenging particularly for a mobile agent. The main challenge is that when a mobile agent try to locate and follow a targeted person, this person can be in a crowd, be occluded by other people, and/or be facing (partially) away from the mobile agent.

130

130

Machine Learning Research at Apple

MARCH 11, 2024

This paper was accepted at The 5th AAAI Workshop on Privacy-Preserving Artificial Intelligence. Personalized recommendations form an important part of today's internet ecosystem, helping artists and creators to reach interested users, and helping users to discover new and engaging content. However, many users today are skeptical of platforms that personalize recommendations, in part due to historically careless treatment of personal data and data privacy.

Algorithm 147

Algorithm Artificial Intelligence Artificial Intelligence 147

Machine Learning Research at Apple

MARCH 10, 2024

As the repository of publicly available pre-trained vision foundation models (VFMs) — such as CLIP, DINOv2, and SAM — grows, users face challenges in storage, memory, and computational efficiency when deploying multiple models concurrently. To address these concerns, we introduce a unique approach that merges the capabilities of multiple VFMs into a single efficient multi-task model.

113

113

Machine Learning Research at Apple

MARCH 10, 2024

Hand gesture recognition is becoming a more prevalent mode of human-computer interaction, especially as cameras proliferate across everyday devices. Despite continued progress in this field, gesture customization is often underexplored. Customization is crucial since it enables users to define and demonstrate gestures that are more natural, memorable, and accessible.

147

147

Machine Learning Research at Apple

MARCH 10, 2024

*=Equal Contributors Personal devices have adopted diverse authentication methods, including biometric recognition and passcodes. In contrast, headphones have limited input mechanisms, depending solely on the authentication of connected devices. We present Moonwalk, a novel method for passive user recognition utilizing the built-in headphone accelerometer.

130

130

Machine Learning Research at Apple

MARCH 4, 2024

Paper abstract: Large-scale web-crawled datasets are fundamental for the success of pre-training vision-language models, such as CLIP. However, the inherent noise and potential irrelevance of web-crawled AltTexts pose challenges in achieving precise image-text alignment. Existing methods utilizing large language models (LLMs) for caption rewriting have shown promise on small, curated datasets like CC3M and CC12M.

147

147

Machine Learning Research at Apple

MARCH 3, 2024

In accordance with the principle of "data minimization," many internet companies are opting to record less data. However, this is often at odds with A/B testing efficacy. For experiments with units with multiple observations, one popular data-minimizing technique is to aggregate data for each unit. However, exact quantile estimation requires the full observation-level data.

147

147

Machine Learning Research at Apple

MARCH 3, 2024

This paper has been accepted to the UniReps Workshop in NeurIPS 2023. Contrastive language image pretraining has become the standard approach for training vision language models. Despite the utility of CLIP visual features as global representations for images, they have limitations when it comes to tasks involving object localization, pixel-level understanding of the image, or 3D perception.

147

147

Machine Learning Research at Apple

FEBRUARY 25, 2024

In-context learning with Large Language Models (LLMs) has emerged as a promising avenue of research in Dialog State Tracking (DST). However, the best-performing in-context learning methods involve retrieving and adding similar examples to the prompt, requiring access to labeled training data. Procuring such training data for a wide range of domains and applications is time-consuming, expensive, and, at times, infeasible.

130

130

Machine Learning Research at Apple

FEBRUARY 25, 2024

This paper was accepted at the workshop HSCMA at ICASSP 2024. Voice triggering (VT) enables users to activate their devices by just speaking a trigger phrase. A front-end system is typically used to perform speech enhancement and/or separation, and produces multiple enhanced and/or separated signals. Since conventional VT systems take only single-channel audio as input, channel selection is performed.

130

130

Machine Learning Research at Apple

FEBRUARY 19, 2024

Large language models (LLMs) have the potential to impact a wide range of creative domains, as exemplified in popular text-to-image generators like DALL·E and Midjourney. However, the application of LLMs to motion-based visual design has not yet been explored and presents novels challenges such as how users might effectively describe motion in natural language.

147

147

Machine Learning Research at Apple

FEBRUARY 11, 2024

Convolution-BatchNorm (ConvBN) blocks are integral components in various computer vision tasks and other domains. A ConvBN block can operate in three modes: Train, Eval, and Deploy. While the Train mode is indispensable for training models from scratch, the Eval mode is suitable for transfer learning and beyond, and the Deploy mode is designed for the deployment of models.

147

147