Timezone: »
Language exhibits structure at different scales, ranging from subwords to words, sentences, paragraphs, and documents. To what extent do deep models capture information at these scales, and can we force them to better capture structure across this hierarchy? We approach this question by focusing on individual neurons, analyzing the behavior of their activations at different timescales. We show that signal processing provides a natural framework for separating structure across scales, enabling us to 1) disentangle scale-specific information in existing embeddings and 2) train models to learn more about particular scales. Concretely, we apply spectral filters to the activations of a neuron across an input, producing filtered embeddings that perform well on part of speech tagging (word-level), dialog speech acts classification (utterance-level), or topic classification (document-level), while performing poorly on the other tasks. We also present a prism layer for training models, which uses spectral filters to constrain different neurons to model structure at different scales. Our proposed BERT + Prism model can better predict masked tokens using long-range context and produces multiscale representations that perform better at utterance- and document-level tasks. Our methods are general and readily applicable to other domains besides language, such as images, audio, and video.
Author Information
Alex Tamkin (Stanford University)
Dan Jurafsky (Stanford University)
Noah Goodman (Stanford University)
More from the Same Authors
-
2021 : DABS: a Domain-Agnostic Benchmark for Self-Supervised Learning »
Alex Tamkin · Vincent Liu · Rongfei Lu · Daniel Fein · Colin Schultz · Noah Goodman -
2021 : Learning to solve complex tasks by growing knowledge culturally across generations »
Michael Tessler · Jason Madeano · Pedro Tsividis · Noah Goodman · Josh Tenenbaum -
2022 Poster: Pile of Law: Learning Responsible Data Filtering from the Law and a 256GB Open-Source Legal Dataset »
Peter Henderson · Mark Krass · Lucia Zheng · Neel Guha · Christopher D Manning · Dan Jurafsky · Daniel Ho -
2022 : Lemma: Bootstrapping High-Level Mathematical Reasoning with Learned Symbolic Abstractions »
Zhening Li · Gabriel Poesia Reis e Silva · Omar Costilla Reyes · Noah Goodman · Armando Solar-Lezama -
2022 : On Rate-Distortion Theory in Capacity-Limited Cognition & Reinforcement Learning »
Dilip Arumugam · Mark Ho · Noah Goodman · Benjamin Van Roy -
2022 : In the ZONE: Measuring difficulty and progression in curriculum generation »
Rose Wang · Jesse Mu · Dilip Arumugam · Natasha Jaques · Noah Goodman -
2022 : MATH-AI: Toward Human-Level Mathematical Reasoning »
Francois Charton · Noah Goodman · Behnam Neyshabur · Talia Ringer · Daniel Selsam -
2022 : Learning Mathematical Reasoning for Education »
Noah Goodman -
2022 : Invited Talk: Noah Goodman »
Noah Goodman -
2022 Poster: Assistive Teaching of Motor Control Tasks to Humans »
Megha Srivastava · Erdem Biyik · Suvir Mirchandani · Noah Goodman · Dorsa Sadigh -
2022 Poster: CLEVRER-Humans: Describing Physical and Causal Events the Human Way »
Jiayuan Mao · Xuelin Yang · Xikun Zhang · Noah Goodman · Jiajun Wu -
2022 Poster: Geoclidean: Few-Shot Generalization in Euclidean Geometry »
Joy Hsu · Jiajun Wu · Noah Goodman -
2022 Poster: Active Learning Helps Pretrained Models Learn the Intended Task »
Alex Tamkin · Dat Nguyen · Salil Deshpande · Jesse Mu · Noah Goodman -
2022 Poster: Foundation Posteriors for Approximate Probabilistic Inference »
Mike Wu · Noah Goodman -
2022 Poster: STaR: Bootstrapping Reasoning With Reasoning »
Eric Zelikman · Yuhuai Wu · Jesse Mu · Noah Goodman -
2022 Poster: Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization? »
Rishi Bommasani · Kathleen A. Creel · Ananya Kumar · Dan Jurafsky · Percy Liang -
2022 Poster: DABS 2.0: Improved Datasets and Algorithms for Universal Self-Supervision »
Alex Tamkin · Gaurab Banerjee · Mohamed Owda · Vincent Liu · Shashank Rammoorthy · Noah Goodman -
2022 Poster: Improving Intrinsic Exploration with Language Abstractions »
Jesse Mu · Victor Zhong · Roberta Raileanu · Minqi Jiang · Noah Goodman · Tim Rocktäschel · Edward Grefenstette -
2021 : Spotlight Talk: Learning to solve complex tasks by growing knowledge culturally across generations »
Noah Goodman · Josh Tenenbaum · Michael Tessler · Jason Madeano -
2021 : Multi-party referential communication in complex strategic games »
Jessica Mankewitz · Veronica Boyce · Brandon Waldon · Georgia Loukatou · Dhara Yu · Jesse Mu · Noah Goodman · Michael C Frank -
2021 Workshop: Meaning in Context: Pragmatic Communication in Humans and Machines »
Jennifer Hu · Noga Zaslavsky · Aida Nematzadeh · Michael Franke · Roger Levy · Noah Goodman -
2021 : Opening remarks »
Jennifer Hu · Noga Zaslavsky · Aida Nematzadeh · Michael Franke · Roger Levy · Noah Goodman -
2021 Poster: Emergent Communication of Generalizations »
Jesse Mu · Noah Goodman -
2021 Poster: Contrastive Reinforcement Learning of Symbolic Reasoning Domains »
Gabriel Poesia · WenXin Dong · Noah Goodman -
2021 Poster: Improving Compositionality of Neural Networks by Decoding Representations to Inputs »
Mike Wu · Noah Goodman · Stefano Ermon -
2021 Panel: The Consequences of Massive Scaling in Machine Learning »
Noah Goodman · Melanie Mitchell · Joelle Pineau · Oriol Vinyals · Jared Kaplan -
2019 : Poster Session »
Ahana Ghosh · Javad Shafiee · Akhilan Boopathy · Alex Tamkin · Theodoros Vasiloudis · Vedant Nanda · Ali Baheri · Paul Fieguth · Andrew Bennett · Guanya Shi · Hao Liu · Arushi Jain · Jacob Tyo · Benjie Wang · Boxiao Chen · Carroll Wainwright · Chandramouli Shama Sastry · Chao Tang · Daniel S. Brown · David Inouye · David Venuto · Dhruv Ramani · Dimitrios Diochnos · Divyam Madaan · Dmitrii Krashenikov · Joel Oren · Doyup Lee · Eleanor Quint · elmira amirloo · Matteo Pirotta · Gavin Hartnett · Geoffroy Dubourg-Felonneau · Gokul Swamy · Pin-Yu Chen · Ilija Bogunovic · Jason Carter · Javier Garcia-Barcos · Jeet Mohapatra · Jesse Zhang · Jian Qian · John Martin · Oliver Richter · Federico Zaiter · Tsui-Wei Weng · Karthik Abinav Sankararaman · Kyriakos Polymenakos · Lan Hoang · mahdieh abbasi · Marco Gallieri · Mathieu Seurin · Matteo Papini · Matteo Turchetta · Matthew Sotoudeh · Mehrdad Hosseinzadeh · Nathan Fulton · Masatoshi Uehara · Niranjani Prasad · Oana-Maria Camburu · Patrik Kolaric · Philipp Renz · Prateek Jaiswal · Reazul Hasan Russel · Riashat Islam · Rishabh Agarwal · Alexander Aldrick · Sachin Vernekar · Sahin Lale · Sai Kiran Narayanaswami · Samuel Daulton · Sanjam Garg · Sebastian East · Shun Zhang · Soheil Dsidbari · Justin Goodwin · Victoria Krakovna · Wenhao Luo · Wesley Chung · Yuanyuan Shi · Yuh-Shyang Wang · Hongwei Jin · Ziping Xu -
2019 Poster: Variational Bayesian Optimal Experimental Design »
Adam Foster · Martin Jankowiak · Elias Bingham · Paul Horsfall · Yee Whye Teh · Thomas Rainforth · Noah Goodman -
2019 Spotlight: Variational Bayesian Optimal Experimental Design »
Adam Foster · Martin Jankowiak · Elias Bingham · Paul Horsfall · Yee Whye Teh · Thomas Rainforth · Noah Goodman -
2018 Poster: Bias and Generalization in Deep Generative Models: An Empirical Study »
Shengjia Zhao · Hongyu Ren · Arianna Yuan · Jiaming Song · Noah Goodman · Stefano Ermon -
2018 Spotlight: Bias and Generalization in Deep Generative Models: An Empirical Study »
Shengjia Zhao · Hongyu Ren · Arianna Yuan · Jiaming Song · Noah Goodman · Stefano Ermon -
2018 Poster: Embedding Logical Queries on Knowledge Graphs »
Will Hamilton · Payal Bajaj · Marinka Zitnik · Dan Jurafsky · Jure Leskovec -
2018 Poster: Multimodal Generative Models for Scalable Weakly-Supervised Learning »
Mike Wu · Noah Goodman -
2017 : Morning panel discussion »
Jürgen Schmidhuber · Noah Goodman · Anca Dragan · Pushmeet Kohli · Dhruv Batra -
2017 : "Language in context" »
Noah Goodman -
2017 Poster: Learning Disentangled Representations with Semi-Supervised Deep Generative Models »
Siddharth Narayanaswamy · Brooks Paige · Jan-Willem van de Meent · Alban Desmaison · Noah Goodman · Pushmeet Kohli · Frank Wood · Philip Torr -
2016 Poster: Neurally-Guided Procedural Models: Amortized Inference for Procedural Graphics Programs using Neural Networks »
Daniel Ritchie · Anna Thomas · Pat Hanrahan · Noah Goodman -
2015 Workshop: Bounded Optimality and Rational Metareasoning »
Samuel J Gershman · Falk Lieder · Tom Griffiths · Noah Goodman -
2013 Poster: Learning and using language via recursive pragmatic reasoning about other agents »
Nathaniel J Smith · Noah Goodman · Michael C Frank -
2013 Poster: Learning Stochastic Inverses »
Andreas Stuhlmüller · Jacob Taylor · Noah Goodman -
2012 Workshop: Probabilistic Programming: Foundations and Applications (2 day) »
Vikash Mansinghka · Daniel Roy · Noah Goodman -
2012 Workshop: Probabilistic Programming: Foundations and Applications (2 day) »
Vikash Mansinghka · Daniel Roy · Noah Goodman -
2012 Poster: Burn-in, bias, and the rationality of anchoring »
Falk Lieder · Tom Griffiths · Noah Goodman -
2011 Poster: Nonstandard Interpretations of Probabilistic Programs for Efficient Inference »
David Wingate · Noah Goodman · Andreas Stuhlmueller · Jeffrey Siskind