Adam Roberts
Adam Roberts earned his PhD in Computer Science from UC Berkeley with a designated emphasis in Computational and Genomic Biology. He has since focused on combining music and technology at Google, where he helped to organize the world's music knowledge for Google Play and is now applying deep learning to the generation of music and art for Google Brain team's Magenta project.
Research Areas
Authored Publications
Sort By
MusicLM: Generating Music From Text
Andrea Agostinelli
Timo Denk
Zalán Borsos
Mauro Verzetti
Antoine Caillon
Qingqing Huang
Marco Tagliasacchi
Matt Sharifi
Neil Zeghidour
Christian Frank
under review (2023)
Preview abstract
We introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption. To support future research, we publicly release MusicCaps, a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts.
Further links: samples, MusicCaps dataset
View details
The Flan Collection: Designing Data and Methods for Effective Instruction Tuning
Shayne Longpre
Le Hou
Albert Webson
Hyung Won Chung
Yi Tay
Denny Zhou
Barret Zoph
Jason Wei
Proceedings of the 40th International Conference on Machine Learning, PMLR (2023), pp. 22631-22648
Preview abstract
We study the design decisions of publicly available instruction tuning methods, and break down the development of Flan 2022 (Chung et al., 2022). Through careful ablation studies on the Flan Collection of tasks and methods, we tease apart the effect of design decisions which enable Flan-T5 to outperform prior work by 3-17%+ across evaluation settings. We find task balancing and enrichment techniques are overlooked but critical to effective instruction tuning, and in particular, training with mixed prompt settings (zero-shot, few-shot, and chain-of-thought) actually yields stronger (2%+) performance in all settings. In further experiments, we show Flan-T5 requires less finetuning to converge higher and faster than T5 on single downstream tasks, motivating instruction-tuned models as more computationally-efficient starting checkpoints for new tasks. Finally, to accelerate research on instruction tuning, we make the Flan 2022 collection of datasets, templates, and methods publicly available at https://github.com/google-research/FLAN/tree/main/flan/v2.
View details
Character-Aware Models Improve Visual Text Rendering
Chitwan Saharia
William Chan
Sharan Narang
Irina Blok
RJ Mical
Mohammad Norouzi
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (2023)
Preview abstract
Current image generation models struggle to reliably produce well-formed visual text. In this paper, we investigate a key contributing factor: popular text-to-image models lack character-level input features, making it much harder to predict a word's visual makeup as a series of glyphs. To quantify this effect, we conduct a series of experiments comparing character-aware vs. character-blind text encoders. In the text-only domain, we find that character-aware models provide large gains on a novel spelling task (WikiSpell). Applying our learnings to the visual domain, we train a suite of image generation models, and show that character-aware variants outperform their character-blind counterparts across a range of novel text rendering tasks (our DrawText benchmark). Our models set a much higher state-of-the-art on visual spelling, with 30+ point accuracy gains over competitors on rare words, despite training on far fewer examples.
View details
LaMDA: Language Models for Dialog Applications
Aaron Daniel Cohen
Alena Butryna
Alicia Jin
Apoorv Kulshreshtha
Ben Zevenbergen
Chung-ching Chang
Cosmo Du
Daniel De Freitas Adiwardana
Dehao Chen
Dmitry (Dima) Lepikhin
Erin Hoffman-John
Igor Krivokon
James Qin
Jamie Hall
Joe Fenton
Johnny Soraker
Kathy Meier-Hellstern
Maarten Paul Bosma
Marc Joseph Pickett
Marcelo Amorim Menegali
Marian Croak
Maxim Krikun
Rachel Bernstein
Ravi Rajakumar
Ray Kurzweil
Romal Thoppilan
Steven Zheng
Taylor Bos
Toju Duke
Tulsee Doshi
Vincent Y. Zhao
Will Rusch
Yanping Huang
Yuanzhong Xu
Zhifeng Chen
arXiv (2022)
Preview abstract
We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and are pre-trained on 1.56T words of public dialog data and web text. While model scaling alone can improve quality, it shows less improvements on safety and factual grounding. We demonstrate that fine-tuning with annotated data and enabling the model to consult external knowledge sources can lead to significant improvements towards the two key challenges of safety and factual grounding.The first challenge, safety, involves ensuring that the model’s responses are consistent with a set of human values, such as preventing harmful suggestions and unfair bias. We quantify safety using a metric based on an illustrative set of values, and we find that filtering candidate responses using aLaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promising approach to improving model safety. The second challenge, factual grounding, involves enabling the model to consult external knowledge sources, such as an information retrieval system, a language translator, and a calculator. We quantify factuality using a groundedness metric, and we find that our approach enables the model to generate responses grounded in known sources, rather than responses that merely sound plausible. Finally, we explore the use of LaMDA in the domains of education and content recommendations, and analyze their helpfulness and role consistency.
View details
PaLM: Scaling Language Modeling with Pathways
Aakanksha Chowdhery
Sharan Narang
Jacob Devlin
Maarten Bosma
Gaurav Mishra
Hyung Won Chung
Sebastian Gehrmann
Parker Schuh
Sasha Tsvyashchenko
Abhishek Rao
Yi Tay
Noam Shazeer
Nan Du
Reiner Pope
James Bradbury
Jacob Austin
Guy Gur-Ari
Pengcheng Yin
Toju Duke
Henryk Michalewski
Xavier Garcia
Liam Fedus
Denny Zhou
David Luan
Barret Zoph
Ryan Sepassi
David Dohan
Shivani Agrawal
Mark Omernick
Andrew M. Dai
Marie Pellat
Aitor Lewkowycz
Erica Moreira
Rewon Child
Oleksandr Polozov
Katherine Lee
Zongwei Zhou
Brennan Saeta
Michele Catasta
Jason Wei
Kathy Meier-Hellstern
arxiv:2204.02311 (2022)
Preview abstract
Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning, which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model PaLM. We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-of-the-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance, outperforming the finetuned state-of-the-art on a suite of multi-step reasoning tasks, and outperforming average human performance on the recently released BIG-bench benchmark. A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model. PaLM also has strong capabilities in multilingual tasks and source code generation, which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity, and study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies.
View details
mT5: A massively multilingual pre-trained text-to-text transformer
Linting Xue
Mihir Sanjay Kale
Rami Al-Rfou
Aditya Siddhant
Aditya Barua
Colin Raffel
Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL 2021), Association for Computational Linguistics, Online, pp. 483-498
Preview abstract
The recent “Text-to-Text Transfer Transformer” (T5) leveraged a unified text-to-text format and scale to attain state-of-the-art results on a wide variety of English-language NLP tasks. In this paper, we introduce mT5, a multilingual variant of T5 that was pre-trained on a new Common Crawl-based dataset covering 101 languages. We detail the design and modified training of mT5 and demonstrate its state-of-the-art performance on many multilingual benchmarks. We also describe a simple technique to prevent “accidental translation” in the zero-shot setting, where a generative model chooses to (partially) translate its prediction into the wrong language. All of the code and model checkpoints used in this work are publicly available.
View details
Preview abstract
Most generative models of audio directly generate samples in one of two domains: time or frequency. While sufficient to express any signal, these representations are inefficient, as they do not utilize existing knowledge of how sound is generated and perceived. A third approach (vocoders/synthesizers) successfully incorporates strong domain knowledge of signal processing and perception, but has been less actively researched due to limited expressivity and difficulty integrating with modern auto-differentiation-based machine learning methods. In this paper, we introduce the Differentiable Digital Signal Processing (DDSP) library, which enables direct integration of classic signal processing elements with deep learning methods. Focusing on audio synthesis, we achieve high-fidelity generation without the need for large autoregressive models or adversarial losses, demonstrating that DDSP enables utilizing strong inductive biases without losing the expressive power of neural networks. Further, we show that combining interpretable modules permits manipulation of each separate model component, with applications such as independent control of pitch and loudness, realistic extrapolation to pitches not seen during training, blind dereverberation of room acoustics, transfer of extracted room acoustics to new environments, and transformation of timbre between disparate sources. In short, DDSP enables an interpretable and modular approach to generative modeling, without sacrificing the benefits of deep learning. The library will be made available upon paper acceptance and we encourage further contributions from the community and domain experts.
View details
Self-supervised Pitch Detection by Inverse Audio Synthesis
Lamtharn (Hanoi) Hantrakul
Rigel Jacob Swavely
Curtis Glenn-Macway Hawthorne
ICML 2020 Workshop on Self-supervision in Audio and Speech (2020) (to appear)
Preview abstract
Audio scene understanding, parsing sound into a hierarchy of meaningful parts, is an open problem in representation learning. Sound is a particularly challenging domain due to its high dimensionality, sequential dependencies and hierarchical structure. Differentiable Digital Signal Processing (DDSP) greatly simplifies the forward problem of generating audio by introducing differentiable synthesizer and effects modules that combine strong signal priors with end-to-end learning. Here, we focus on the inverse problem, inferring synthesis parameters to approximate an audio scene. We demonstrate that DDSP modules can enable a new approach to self-supervision, generating synthetic audio with differentiable synthesizers and training feature extractor networks to infer the synthesis parameters. By building a hierarchy from sinusoidal to harmonic representations, we show that it possible to use such an inverse modeling approach to disentangle pitch from timbre, an important task in audio scene understanding.
View details
Identifying the intersections: User experience + research scientist collaboration in a generative machine learning interface
Jess Scon Holbrook
ACM CHI Conference 2019 (2019)
Preview abstract
Creative generative machine learning interfaces are stronger when multiple actors bearing different points of view actively contribute to them. User experience (UX) research and design involvement in the creation of machine learning (ML) models help ML research scientists to more effectively identify human needs that ML models will fulfill. The People and AI Research (PAIR) group within Google developed a novel program method in which UXers are embedded into an ML research group for three months to provide a human-centered perspective on the creation of ML models. The first full-time cohort of UXers were embedded in a team of ML research scientists focused on deep generative models to assist in music composition. Here, we discuss the structure and goals of the program, challenges we faced during execution, and insights gained as a result of the process. We offer practical suggestions for how to foster communication between UX and ML research teams and recommended UX design processes for building creative generative machine learning interfaces.
View details
Enabling Factorized Piano Music Modeling and Generation with the MAESTRO Dataset
Curtis Hawthorne
Andrew Stasyuk
Sander Dieleman
Erich Elsen
ICLR (2019)
Preview abstract
Generating musical audio directly with neural networks is notoriously difficult because it requires coherently modeling both long- and short-term structure. Fortunately, most music is also highly structured and primarily composed of discrete note events played on musical instruments. Herein, we show that by using notes as an intermediate representation, we can train a suite of models capable of transcribing, composing, and synthesizing audio waveforms with coherent musical structure on timescales spanning six orders of magnitude (~0.01 ms (8 kHz) to ~100 s). This large advance in the state of the art is enabled by our release of the new MAESTRO (MIDI and Audio Edited for Synchronous TRacks and Organization) dataset, composed of over 172 hours of virtuosic piano performances captured with fine alignment (~3 ms) between note labels and audio waveforms. The networks and the dataset together present a promising approach toward creating new expressive and interpretable neural models of music.
View details
