title
stringlengths
14
150
paper_url
stringlengths
42
42
authors
listlengths
1
28
type
stringclasses
3 values
primary_area
stringclasses
0 values
abstract
large_stringlengths
386
2.56k
keywords
listlengths
1
18
TL;DR
large_stringlengths
0
250
submission_number
int64
8
15.6k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
embedding
listlengths
768
768
Online PCA in Converging Self-consistent Field Equations
https://openreview.net/forum?id=vq11gurmUY
[ "Xihan Li", "Xiang Chen", "Rasul Tutunov", "Haitham Bou Ammar", "Lei Wang", "Jun Wang" ]
Poster
null
Self-consistent Field (SCF) equation is a type of nonlinear eigenvalue problem in which the matrix to be eigen-decomposed is a function of its own eigenvectors. It is of great significance in computational science for its connection to the Schrödinger equation. Traditional fixed-point iteration methods for solving such...
[ "Self-consistent Field Equation", "Computational Science", "Online PCA" ]
We developed a new algorithm based on online PCA to converge Self-consistent Field Equations
15,599
null
null
[ -0.0014557929243892431, -0.0370333157479763, 0.01890290156006813, 0.058369580656290054, 0.02634771727025509, -0.00993984192609787, 0.002929126378148794, 0.025538770481944084, -0.017489392310380936, -0.04744797945022583, 0.02338443137705326, -0.026369599625468254, -0.07381287962198257, 0.03...
Don’t blame Dataset Shift! Shortcut Learning due to Gradients and Cross Entropy
https://openreview.net/forum?id=zyZkaqNnpa
[ "Aahlad Manas Puli", "Lily H Zhang", "Yoav Wald", "Rajesh Ranganath" ]
Poster
null
Common explanations for shortcut learning assume that the shortcut improves prediction only under the training distribution. Thus, models trained in the typical way by minimizing log-loss using gradient descent, which we call default-ERM, should utilize the shortcut. However, even when the stable feature determines the...
[ "shortcut learning", "spurious correlations", "perfect stable feature", "perception tasks", "implicit bias in optimization", "improving inductive biases" ]
Implicit biases toward maximizing margins induce shortcut learning in ERM even in tasks with perfect stable features, controlling margins mitigates shortcuts
15,594
2308.12553
title_snapshot
[ -0.02684474177658558, 0.010417930781841278, 0.014150913804769516, 0.033776361495256424, 0.029840683564543724, 0.02494463510811329, 0.0460018590092659, 0.008711917325854301, -0.036288194358348846, -0.04169975593686104, -0.03859865292906761, 0.05596565827727318, -0.07221315056085587, -0.0319...
On Slicing Optimality for Mutual Information
https://openreview.net/forum?id=JMuKfZx2xU
[ "Ammar Fayad", "Majd Ibrahim" ]
Poster
null
Measuring dependence between two random variables is of great importance in various domains but is difficult to compute in today's complex environments with high-dimensional data. Recently, slicing methods have shown to be a scalable approach to measuring mutual information (MI) between high-dimensional variables by pr...
[ "Mutual information", "Information Theory" ]
null
15,586
null
null
[ -0.02603287622332573, -0.0008994860108941793, -0.0011620805598795414, 0.03221255913376808, 0.036564625799655914, 0.018487419933080673, 0.030353151261806488, -0.014197097159922123, -0.009966477751731873, -0.035363174974918365, -0.015519496984779835, 0.002838997170329094, -0.06200432777404785,...
k-Median Clustering via Metric Embedding: Towards Better Initialization with Differential Privacy
https://openreview.net/forum?id=9zV2OXCrVF
[ "Chenglin Fan", "Ping Li", "Xiaoyun Li" ]
Poster
null
In clustering algorithms, the choice of initial centers is crucial for the quality of the learned clusters. We propose a new initialization scheme for the $k$-median problem in the general metric space (e.g., discrete space induced by graphs), based on the construction of metric embedding tree structure of the data. We...
[ "privacy", "clustering" ]
null
15,580
2206.12895
title_snapshot
[ -0.00804523378610611, 0.004351302515715361, -0.006113056093454361, 0.059019990265369415, 0.04855501651763916, 0.04439820721745491, 0.030489491298794746, -0.017465949058532715, -0.005297380965203047, -0.038149360567331314, -0.013853426091372967, -0.05257459729909897, -0.0398796908557415, -0...
Information Maximization Perspective of Orthogonal Matching Pursuit with Applications to Explainable AI
https://openreview.net/forum?id=CAF4CnUblx
[ "Aditya Chattopadhyay", "Ryan Pilgrim", "Rene Vidal" ]
Spotlight
null
Information Pursuit (IP) is a classical active testing algorithm for predicting an output by sequentially and greedily querying the input in order of information gain. However, IP is computationally intensive since it involves estimating mutual information in high-dimensional spaces. This paper explores Orthogonal Matc...
[ "Information Maximization", "Sparse Coding", "Orthogonal Matching Pursuit", "Explainable AI", "Information Pursuit" ]
We show that the popular OMP algorithm can be derived from information-theoretic principles modulo a normalization factor. We then use this insight to design a computationally simple sparse-coding based explainable AI algorithm.
15,576
null
null
[ -0.04869089275598526, -0.0010465370723977685, 0.0023260130546987057, 0.06991083174943924, 0.04575653374195099, 0.03701650723814964, 0.014841753989458084, 0.00716420216485858, -0.035081859678030014, -0.04167557135224342, -0.024628140032291412, 0.015593503601849079, -0.06310165673494339, -0....
STEVE-1: A Generative Model for Text-to-Behavior in Minecraft
https://openreview.net/forum?id=YkBDJWerKg
[ "Shalev Lifshitz", "Keiran Paster", "Harris Chan", "Jimmy Ba", "Sheila A. McIlraith" ]
Spotlight
null
Constructing AI models that respond to text instructions is challenging, especially for sequential decision-making tasks. This work introduces a methodology, inspired by unCLIP, for instruction-tuning generative models of behavior without relying on a large dataset of instruction-labeled trajectories. Using this method...
[ "minecraft", "instruction following", "foundation models", "sequence models", "reinforcement learning", "sequential decision making", "goal conditioned reinforcement learning", "text conditioned reinforcement learning", "transformers", "deep learning" ]
We introduce a methodology for instruction-tuning generative models of behavior without relying on a large dataset of instruction-labeled trajectories.
15,575
2306.00937
title_snapshot
[ -0.0015651540597900748, 0.004793412983417511, -0.022066764533519745, 0.036189865320920944, 0.033577483147382736, 0.011574724689126015, 0.028844140470027924, 0.02173878811299801, -0.02761303260922432, -0.038901619613170624, -0.03148203343153, 0.014052476733922958, -0.06032954156398773, -0.0...
AMAG: Additive, Multiplicative and Adaptive Graph Neural Network For Forecasting Neuron Activity
https://openreview.net/forum?id=7ntI4kcoqG
[ "Jingyuan Li", "Leo Scholl", "Trung Le", "Pavithra Rajeswaran", "Amy L Orsborn", "Eli Shlizerman" ]
Poster
null
Latent Variable Models (LVMs) propose to model the dynamics of neural populations by capturing low-dimensional structures that represent features involved in neural activity. Recent LVMs are based on deep learning methodology where a deep neural network is trained to reconstruct the same neural activity given as input ...
[ "Neuroscience and Cognitive Science", "Neural Activity Forecasting", "Graph Neural Network" ]
In this work, we emphasize the importance of modeling neural population dynamics via forecasting tasks and aim to improve the forecasting performance with Graph Neural Networks
15,562
null
null
[ -0.016840733587741852, 0.0012876068940386176, 0.008209534920752048, 0.020538803189992905, 0.03522329777479172, 0.02475191466510296, 0.048422615975141525, 0.011483663693070412, -0.09987568855285645, -0.0515366829931736, 0.012643805705010891, -0.010690320283174515, -0.07200940698385239, 0.01...
Conditional Matrix Flows for Gaussian Graphical Models
https://openreview.net/forum?id=GYnbubCXhE
[ "Marcello Massimo Negri", "Fabricio Arend Torres", "Volker Roth" ]
Poster
null
Studying conditional independence among many variables with few observations is a challenging task. Gaussian Graphical Models (GGMs) tackle this problem by encouraging sparsity in the precision matrix through $l_q$ regularization with $q\leq1$. However, most GMMs rely on the $l_1$ norm because the objective is highly n...
[ "normalizing flow", "variational inference", "graphical lasso", "gaussian graphical model", "bayesian inference" ]
General framework for variational inference with matrix-variate Normalizing Flow in Gaussian Graphical Models
15,533
2306.07255
title_snapshot
[ -0.01246718317270279, -0.01000144798308611, 0.007756223436444998, 0.029787471517920494, 0.021703660488128662, 0.04986925423145294, 0.024796312674880028, 0.0013303677551448345, -0.01412421464920044, -0.05236733332276344, -0.020433856174349785, 0.004004163201898336, -0.07325607538223267, -0....
Representational Strengths and Limitations of Transformers
https://openreview.net/forum?id=36DxONZ9bA
[ "Clayton Sanford", "Daniel Hsu", "Matus Telgarsky" ]
Poster
null
Attention layers, as commonly used in transformers, form the backbone of modern deep learning, yet there is no mathematical description of their benefits and deficiencies as compared with other architectures. In this work we establish both positive and negative results on the representation power of attention layers, w...
[ "self-attention", "approximation theory", "communication complexity" ]
We give function approximation tasks that demonstrate the advantages of transformers over RNNs and FNNs, the impact of the self-attention embedding dimension on expressivity, and the limitations of any bounded-size layer of self-attention.
15,514
2306.02896
title_snapshot
[ -0.0075914994813501835, -0.029937664046883583, 0.012856494635343552, 0.02158329077064991, 0.01688380353152752, 0.025267554447054863, 0.017064418643712997, -0.003978664521127939, -0.040782541036605835, -0.03176536411046982, 0.0013909790432080626, 0.009064736776053905, -0.06343846768140793, ...
Cappy: Outperforming and Boosting Large Multi-Task LMs with a Small Scorer
https://openreview.net/forum?id=Srt1hhQgqa
[ "Bowen Tan", "Yun Zhu", "Lijuan Liu", "Eric Xing", "Zhiting Hu", "Jindong Chen" ]
Poster
null
Large language models (LLMs) such as T0, FLAN, and OPT-IML excel in multi-tasking under a unified instruction-following paradigm, where they also exhibit remarkable generalization abilities to unseen tasks. Despite their impressive performance, these LLMs, with sizes ranging from several billion to hundreds of billions...
[ "multi-task", "large language models", "pretrain model" ]
we introduce a pretrained small scorer, Cappy, designed to enhance the performance and efficiency of multi-task LLMs.
15,503
2311.06720
title_snapshot
[ -0.009453810751438141, -0.019826430827379227, 0.009188784286379814, 0.011240563355386257, 0.03876345232129097, 0.003912144340574741, -0.006743809673935175, 0.006919141858816147, -0.022162489593029022, -0.003882643999531865, -0.024423886090517044, 0.04255705326795578, -0.06767383217811584, ...
Two-Stage Learning to Defer with Multiple Experts
https://openreview.net/forum?id=GIlsH0T4b2
[ "Anqi Mao", "Christopher Mohri", "Mehryar Mohri", "Yutao Zhong" ]
Poster
null
We study a two-stage scenario for learning to defer with multiple experts, which is crucial in practice for many applications. In this scenario, a predictor is derived in a first stage by training with a common loss function such as cross-entropy. In the second stage, a deferral function is learned to assign the most...
[ "learning to defer", "learning theory" ]
null
15,489
null
null
[ -0.01720811054110527, -0.02738632634282112, 0.017594238743185997, 0.02812323346734047, 0.022113636136054993, 0.020020311698317528, 0.04065193980932236, -0.009395431727170944, -0.007776045706123114, -0.023939223960042, 0.01157231442630291, 0.03940389305353165, -0.03154143691062927, -0.01582...
Multiply Robust Federated Estimation of Targeted Average Treatment Effects
https://openreview.net/forum?id=M6UccKMFGl
[ "Larry Han", "Zhu Shen", "Jose R Zubizarreta" ]
Poster
null
Federated or multi-site studies have distinct advantages over single-site studies, including increased generalizability, the ability to study underrepresented populations, and the opportunity to study rare exposures and outcomes. However, these studies are complicated by the need to preserve the privacy of each individ...
[ "Causal inference", "Covariate mismatch", "Federated learning", "Multiple robustness", "Transportation" ]
We propose a federated approach that enables valid causal inferences for a target population using multi-site data, accounting for covariate shift and covariate mismatch through privacy-preserving and multiply robust nuisance function estimation.
15,483
2309.12600
title_snapshot
[ 0.0061089093796908855, -0.05090666934847832, 0.011144788935780525, 0.014441552571952343, 0.04095268249511719, 0.027548793703317642, 0.036499448120594025, -0.030560093000531197, 0.005949575919657946, -0.06086837872862816, 0.05730157345533371, -0.03224799036979675, -0.0458882637321949, -0.00...
On the Variance, Admissibility, and Stability of Empirical Risk Minimization
https://openreview.net/forum?id=4KV2xLeqPN
[ "Gil Kur", "Eli Putterman", "Alexander Rakhlin" ]
Spotlight
null
It is well known that Empirical Risk Minimization (ERM) may attain minimax suboptimal rates in terms of the mean squared error (Birgé and Massart, 1993). In this paper, we prove that, under relatively mild assumptions, the suboptimality of ERM must be due to its bias. Namely, the variance error term of ERM (in terms of...
[ "empirical risk minimization", "bias-variance decomposition", "admissibility" ]
We show that the variance of Empirical Risk Minimization enjoys the minimax rate in various settings.
15,480
2305.18508
title_snapshot
[ -0.035826247185468674, -0.004615152720361948, 0.007494011428207159, 0.03763628005981445, 0.04493910074234009, 0.052696261554956436, 0.014183620922267437, -0.01014759298413992, -0.018169980496168137, -0.062185101211071014, 0.0008036557701416314, 0.004955780692398548, -0.053135763853788376, ...
To Stay or Not to Stay in the Pre-train Basin: Insights on Ensembling in Transfer Learning
https://openreview.net/forum?id=NNooZoQpP4
[ "Ildus Sadrtdinov", "Dmitrii Pozdeev", "Dmitry P. Vetrov", "Ekaterina Lobacheva" ]
Poster
null
Transfer learning and ensembling are two popular techniques for improving the performance and robustness of neural networks. Due to the high cost of pre-training, ensembles of models fine-tuned from a single pre-trained checkpoint are often used in practice. Such models end up in the same basin of the loss landscape,...
[ "ensembles", "transfer learning", "loss landscape basins", "model soups" ]
We study the effectiveness of the exploration of the pre-train basin and its close vicinity for ensembling in transfer learning.
15,466
2303.03374
title_snapshot
[ -0.03074067085981369, -0.03417517989873886, -0.029967624694108963, 0.04366062209010124, 0.07486872375011444, 0.021109197288751602, 0.024840405210852623, -0.009222251363098621, -0.02142038382589817, -0.04807858169078827, -0.021968437358736992, 0.004244836047291756, -0.04492251202464104, -0....
Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task
https://openreview.net/forum?id=frVo9MzRuU
[ "Maya Okawa", "Ekdeep Singh Lubana", "Robert P. Dick", "Hidenori Tanaka" ]
Poster
null
Modern generative models exhibit unprecedented capabilities to generate extremely realistic data. However, given the inherent compositionality of the real world, reliable use of these models in practical applications requires that they exhibit the capability to compose a novel set of concepts to generate outputs not se...
[ "Diffusion model; Emergence; Emergent capabilities; Science of deep learning; Mechanistic interpretability" ]
null
15,432
2310.09336
title_snapshot
[ -0.011007788591086864, -0.02262227237224579, -0.015081789344549179, 0.06097687408328056, 0.06146162748336792, 0.008193667978048325, 0.019745508208870888, 0.023355983197689056, -0.010946705006062984, -0.026890484616160393, -0.018293533474206924, -0.009558286517858505, -0.06979892402887344, ...
Phase diagram of early training dynamics in deep neural networks: effect of the learning rate, depth, and width
https://openreview.net/forum?id=Al9yglQGKj
[ "Dayal Singh Kalra", "Maissam Barkeshli" ]
Poster
null
We systematically analyze optimization dynamics in deep neural networks (DNNs) trained with stochastic gradient descent (SGD) and study the effect of learning rate $\eta$, depth $d$, and width $w$ of the neural network. By analyzing the maximum eigenvalue $\lambda^H_t$ of the Hessian of the loss, which is a measure of ...
[ "Optimization dynamics", "Phase diagrams", "learning rate transition", "Catapult effect" ]
null
15,410
2302.12250
title_snapshot
[ -0.04206845909357071, -0.020078321918845177, -0.02125728316605091, 0.03637029230594635, 0.03919079899787903, 0.010505006648600101, 0.021756136789917946, 0.011290469206869602, -0.030940616503357887, -0.0133463554084301, -0.010869750753045082, 0.00915936753153801, -0.029022041708230972, 0.02...
Explaining V1 Properties with a Biologically Constrained Deep Learning Architecture
https://openreview.net/forum?id=1uirUsR9E7
[ "Galen Pogoncheff", "Jacob Granley", "Michael Beyeler" ]
Poster
null
Convolutional neural networks (CNNs) have recently emerged as promising models of the ventral visual stream, despite their lack of biological specificity. While current state-of-the-art models of the primary visual cortex (V1) have surfaced from training with adversarial examples and extensively augmented data, these m...
[ "NeuroAI", "Neuroscience", "Visual Stream", "Convolutional Neural Networks", "Biologically inspired deep learning" ]
null
15,381
2305.11275
title_snapshot
[ -0.011736673302948475, -0.002720854477956891, 0.006407939828932285, 0.01658070646226406, 0.014261783100664616, 0.02526254579424858, 0.021855764091014862, 0.016574693843722343, -0.037679318338632584, -0.046442221850156784, -0.01620679534971714, -0.014867845922708511, -0.06232528015971184, 0...
Adversarial Examples Might be Avoidable: The Role of Data Concentration in Adversarial Robustness
https://openreview.net/forum?id=JDoA6admhv
[ "Ambar Pal", "Jeremias Sulam", "Rene Vidal" ]
Poster
null
The susceptibility of modern machine learning classifiers to adversarial examples has motivated theoretical results suggesting that these might be unavoidable. However, these results can be too general to be applicable to natural data distributions. Indeed, humans are quite robust for tasks involving vision. This appar...
[ "Adversarial Robustness", "Geometry in Data", "Low Dimensional Modeling" ]
Adversarial examples are provably avoidable when data distributions are concentrated
15,380
2309.16096
title_snapshot
[ -0.009013594128191471, -0.014007702469825745, -0.007669204846024513, 0.04679276794195175, 0.03666902706027031, 0.004623149987310171, 0.029620366171002388, -0.031539253890514374, -0.02135789953172207, -0.027203671634197235, -0.021058067679405212, -0.014150319620966911, -0.07849422097206116, ...
Improving *day-ahead* Solar Irradiance Time Series Forecasting by Leveraging Spatio-Temporal Context
https://openreview.net/forum?id=x5ZruOa4ax
[ "Oussama Boussif", "Ghait Boukachab", "Dan Assouline", "Stefano Massaroli", "Tianle Yuan", "Loubna Benabbou", "Yoshua Bengio" ]
Poster
null
Solar power harbors immense potential in mitigating climate change by substantially reducing CO$_{2}$ emissions. Nonetheless, the inherent variability of solar irradiance poses a significant challenge for seamlessly integrating solar power into the electrical grid. While the majority of prior research has centered on e...
[ "Time series forecasting", "multi-modal learning", "solar irradiance", "context-enriched learning" ]
We present a novel deep learning architecture which leverages the spatio-temporal context (e.g. satellite data) around a meteorological station to improve the forecasting of solar irradiance at this station.
15,362
2306.01112
title_snapshot
[ 0.0013459994224831462, -0.04608852416276932, -0.009319248609244823, 0.025752831250429153, 0.04123653843998909, 0.01641872525215149, 0.0357096903026104, 0.023305663838982582, -0.03409735858440399, -0.04727848619222641, -0.0059935241006314754, -0.008005003444850445, -0.06459356099367142, 0.0...
Red Teaming Deep Neural Networks with Feature Synthesis Tools
https://openreview.net/forum?id=Od6CHhPM7I
[ "Stephen Casper", "Tong Bu", "Yuxiao Li", "Jiawei Li", "Kevin Zhang", "Kaivalya Hariharan", "Dylan Hadfield-Menell" ]
Poster
null
Interpretable AI tools are often motivated by the goal of understanding model behavior in out-of-distribution (OOD) contexts. Despite the attention this area of study receives, there are comparatively few cases where these tools have identified previously unknown bugs in models. We argue that this is due, in part, to a...
[ "interpretability", "benchmarking", "auditing", "diagnostics", "debugging", "adversarial attacks", "feature synthesis" ]
We use trojan discovery as an evaluation task to study how useful interpretability tools are for model debugging.
15,355
2302.10894
title_snapshot
[ -0.005792642943561077, -0.025583703070878983, -0.04953250288963318, 0.05164186283946037, 0.03521556034684181, 0.010797987692058086, 0.010742461308836937, 0.0022777088452130556, -0.022295355796813965, -0.028810853138566017, -0.021500378847122192, 0.032621607184410095, -0.0679154172539711, -...
From Pixels to UI Actions: Learning to Follow Instructions via Graphical User Interfaces
https://openreview.net/forum?id=3PjCt4kmRx
[ "Peter Shaw", "Mandar Joshi", "James Cohan", "Jonathan Berant", "Panupong Pasupat", "Hexiang Hu", "Urvashi Khandelwal", "Kenton Lee", "Kristina Toutanova" ]
Spotlight
null
Much of the previous work towards digital agents for graphical user interfaces (GUIs) has relied on text-based representations (derived from HTML or other structured data sources), which are not always readily available. These input representations have been often coupled with custom, task-specific action spaces. This...
[ "instruction following", "web tasks", "user interface tasks", "vision and language", "representation learning", "reinforcement learning", "imitation learning", "tree search", "language grounding", "web agents", "computer control" ]
We study GUI-based instruction following with a general observation and action space consisting of pixel-based inputs and low-level actions.
15,350
2306.00245
title_snapshot
[ -0.009596582502126694, -0.01779075898230076, -0.04162798449397087, 0.016404440626502037, 0.016839314252138138, -0.0016211661277338862, 0.03397488221526146, 0.045159488916397095, 0.0035678294952958822, -0.03607936576008797, -0.04605865851044655, 0.033142536878585815, -0.07055862247943878, -...
Human-in-the-Loop Optimization for Deep Stimulus Encoding in Visual Prostheses
https://openreview.net/forum?id=ZED5wdGous
[ "Jacob Granley", "Tristan Fauvel", "Matthew Chalk", "Michael Beyeler" ]
Poster
null
Neuroprostheses show potential in restoring lost sensory function and enhancing human capabilities, but the sensations produced by current devices often seem unnatural or distorted. Exact placement of implants and differences in individual perception lead to significant variations in stimulus response, making personali...
[ "Brain Computer Interfaces", "BCI", "Stimulus Encoding", "Visual Prostheses", "Bayesian Optimization", "Preferential Bayesian Optimization", "Human-in-the-loop Optimization", "Sensory Neuroprostheses", "Neuroprostheses", "Patient-Specific Optimization", "Latent Space Bayesian Optimization" ]
We use deep stimulus encoding combined with Bayesian optimization to personalize stimulation strategies in visual prostheses.
15,347
2306.13104
title_snapshot
[ 0.009315786883234978, 0.02004309929907322, 0.0027953863609582186, 0.010183311067521572, 0.06479554623365402, 0.052021708339452744, 0.02349875308573246, 0.007941216230392456, -0.02671494521200657, -0.04795370623469353, -0.00800853781402111, 0.016003094613552094, -0.02907443419098854, -0.017...
Convex-Concave Zero-Sum Markov Stackelberg Games
https://openreview.net/forum?id=0rEJx5QAxt
[ "Denizalp Goktas", "Arjun Prakash", "Amy Greenwald" ]
Poster
null
Zero-sum Markov Stackelberg games can be used to model myriad problems, in domains ranging from economics to human robot interaction. In this paper, we develop policy gradient methods that solve these games in continuous state and action settings using noisy gradient estimates computed from observed trajectories of pla...
[ "Stackelberg games", "Equilibrium Computation", "Policy Gradient" ]
We develop policy gradient methods to solve zero-sum stochastic Stackelberg games from observed play, and apply our methods to solve reach avoid problems.
15,332
2401.12437
title_snapshot
[ -0.041312187910079956, 0.012923473492264748, -0.005955120082944632, 0.04206971824169159, 0.051817554980516434, 0.033532772213220596, 0.017266003414988518, 0.02920207753777504, -0.04297052323818207, -0.038914136588573456, -0.006318268831819296, 0.004925713874399662, -0.07700027525424957, -0...
Agnostically Learning Single-Index Models using Omnipredictors
https://openreview.net/forum?id=frHPeRedHo
[ "Aravind Gollakota", "Parikshit Gopalan", "Adam Klivans", "Konstantinos Stavropoulos" ]
Poster
null
We give the first result for agnostically learning Single-Index Models (SIMs) with arbitrary monotone and Lipschitz activations. All prior work either held only in the realizable setting or required the activation to be known. Moreover, we only require the marginal to have bounded second moments, whereas all prior work...
[ "generalized linear models", "single-index models", "agnostic learning", "pac learning", "logistic regression", "omnipredictors", "multiaccuracy", "calibration" ]
null
15,293
2306.10615
title_snapshot
[ -0.015321166254580021, -0.007166577037423849, 0.02172389067709446, 0.05035260319709778, 0.04708294942975044, 0.05254729837179184, 0.03275302052497864, -0.002424070378765464, -0.046119969338178635, -0.02696705050766468, 0.0014232993125915527, 0.017009172588586807, -0.08330706506967545, -0.0...
Training Chain-of-Thought via Latent-Variable Inference
https://openreview.net/forum?id=a147pIS2Co
[ "Du Phan", "Matthew Douglas Hoffman", "david dohan", "Sholto Douglas", "Tuan Anh Le", "Aaron T Parisi", "Pavel Sountsov", "Charles Sutton", "Sharad Vikram", "Rif A. Saurous" ]
Poster
null
Large language models (LLMs) solve problems more accurately and interpretably when instructed to work out the answer step by step using a "chain-of-thought" (CoT) prompt. One can also improve LLMs' performance on a specific task by supervised fine-tuning, i.e., by using gradient ascent on some tunable parameters to max...
[ "Large language models", "latent-variable models", "control variates", "chain-of-thought", "MCMC" ]
Treating chain-of-thought-prompted question-answering LLMs as probabilistic latent-variable models lets us derive a principled, simple, effective way of tuning them to generate better rationales without training on human-generated rationales.
15,280
2312.02179
title_snapshot
[ -0.016337187960743904, -0.014040770940482616, -0.007466981187462807, 0.058216553181409836, 0.05161978676915169, 0.01831459254026413, 0.03033926896750927, -0.004583459813147783, -0.016107449308037758, 0.002620916347950697, -0.01807704195380211, 0.044307537376880646, -0.06870155781507492, -0...
Combining Behaviors with the Successor Features Keyboard
https://openreview.net/forum?id=GhNCFtLSsy
[ "Wilka Carvalho", "Andre Saraiva", "Angelos Filos", "Andrew Kyle Lampinen", "Loic Matthey", "Richard Lewis", "Honglak Lee", "Satinder Singh", "Danilo Jimenez Rezende", "Daniel Zoran" ]
Poster
null
The Option Keyboard (OK) was recently proposed as a method for transferring behavioral knowledge across tasks. OK transfers knowledge by adaptively combining subsets of known behaviors using Successor Features (SFs) and Generalized Policy Improvement (GPI). However, it relies on hand-designed state-features and task en...
[ "deep reinforcement learning", "successor features", "transfer", "generalization", "feature-discovery" ]
We are the first method to enable transfer with successor features in a complex 3D environment when all needed representations are discovered
15,278
2310.15940
title_snapshot
[ -0.034866586327552795, -0.029159031808376312, 0.015153174288570881, 0.009531697258353233, 0.06730277091264725, 0.03062564693391323, 0.024724341928958893, -0.014228030107915401, -0.026518147438764572, -0.04931975156068802, -0.007761197630316019, -0.0013439774047583342, -0.07278537005186081, ...
Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation
https://openreview.net/forum?id=NnMEadcdyD
[ "Diederik P Kingma", "Ruiqi Gao" ]
Oral
null
To achieve the highest perceptual quality, state-of-the-art diffusion models are optimized with objectives that typically look very different from the maximum likelihood and the Evidence Lower Bound (ELBO) objectives. In this work, we reveal that diffusion model objectives are actually closely related to the ELBO. Spe...
[ "Diffusion Model", "Evidence Lower Bound", "Maximum Likelihood" ]
We develop a theoretical understanding of the training objective of diffusion models as a the ELBOs subject to data augmentation.
15,266
2303.00848
title_snapshot
[ -0.01085861399769783, -0.009492375887930393, 0.018442293629050255, 0.05610140040516853, 0.027521798387169838, 0.019965095445513725, 0.008105548098683357, -0.007711956277489662, -0.027005977928638458, -0.06528965383768082, -0.014215612784028053, -0.010500023141503334, -0.05066535249352455, ...
Language Models Meet World Models: Embodied Experiences Enhance Language Models
https://openreview.net/forum?id=SVBR6xBaMl
[ "Jiannan Xiang", "Tianhua Tao", "Yi Gu", "Tianmin Shu", "Zirui Wang", "Zichao Yang", "Zhiting Hu" ]
Poster
null
While large language models (LMs) have shown remarkable capabilities across numerous tasks, they often struggle with simple reasoning and planning in physical environments, such as understanding object permanence or planning household activities. The limitation arises from the fact that LMs are trained only on written ...
[ "Language Model", "World Model", "Embodied Experience" ]
null
15,264
2305.10626
title_snapshot
[ -0.011420771479606628, -0.007310894783586264, -0.0019524100935086608, 0.0239554513245821, 0.03693156689405441, 0.021153664216399193, 0.028541242703795433, 0.03108379617333412, -0.04716509208083153, -0.01198013499379158, -0.034894220530986786, 0.011222537606954575, -0.0483768992125988, -0.0...
Smooth Flipping Probability for Differential Private Sign Random Projection Methods
https://openreview.net/forum?id=GEMHw2sd9S
[ "Ping Li", "Xiaoyun Li" ]
Poster
null
We develop a series of differential privacy (DP) algorithms from a family of random projection (RP) and sign random projection (SignRP) methods. We first show how to improve the previous DP-RP approach using the ``optimal Gaussian mechanism''. Then, we propose a series of DP-SignRP algorithms that leverage the robustne...
[ "Differential Privacy", "Random Projection" ]
null
15,259
null
null
[ 0.006278572138398886, -0.005695246625691652, 0.026634754613041878, 0.05669298768043518, 0.03331264853477478, 0.03625204414129257, 0.0324786901473999, -0.02870195172727108, -0.023863768205046654, -0.038923799991607666, 0.004661723971366882, -0.020801613107323647, -0.06129138916730881, -0.01...
Data Market Design through Deep Learning
https://openreview.net/forum?id=sgCrNMOuXp
[ "Sai Srivatsa Ravindranath", "Yanchen Jiang", "David C. Parkes" ]
Poster
null
The _data market design_ problem is a problem in economic theory to find a set of signaling schemes (statistical experiments) to maximize expected revenue to the information seller, where each experiment reveals some of the information known to a seller and has a corresponding price. Each buyer has their own decision ...
[ "Data Markets", "Information Design", "Differentiable Economics", "Economics", "Deep Learning", "Mechanism Design", "Algorithmic Game Theory" ]
null
15,244
2310.20096
title_snapshot
[ -0.03054381161928177, -0.02929159812629223, -0.0361216701567173, 0.05858469754457474, 0.05025386065244675, 0.033263664692640305, -0.007914029061794281, 0.004537082742899656, 0.018119893968105316, -0.03043198399245739, -0.025361469015479088, 0.019118735566735268, -0.04152938351035118, -0.00...
Text Alignment Is An Efficient Unified Model for Massive NLP Tasks
https://openreview.net/forum?id=xkkBFePoFn
[ "Yuheng Zha", "Yichi Yang", "Ruichen Li", "Zhiting Hu" ]
Poster
null
Large language models (LLMs), typically designed as a function of next-word prediction, have excelled across extensive NLP tasks. Despite the generality, next-word prediction is often not an efficient formulation for many of the tasks, demanding an extreme scale of model parameters (10s or 100s of billions) and sometim...
[ "Text Alignment", "Efficient Unified Model", "NLU Tasks", "Factual Consistency Evaluation", "QA with Unanswerable Question" ]
null
15,239
2307.02729
title_snapshot
[ 0.006708015687763691, -0.03615410253405571, -0.01540782954543829, 0.05766846239566803, 0.00841592624783516, 0.030705109238624573, 0.019481468945741653, 0.04080408811569214, -0.01430776808410883, 0.0005645148921757936, -0.02500685304403305, 0.04418336972594261, -0.08508304506540298, -0.0396...
When Can We Track Significant Preference Shifts in Dueling Bandits?
https://openreview.net/forum?id=LjWJLkSpjh
[ "Joe Suk", "Arpit Agarwal" ]
Poster
null
The $K$-armed dueling bandits problem, where the feedback is in the form of noisy pairwise preferences, has been widely studied due its applications in information retrieval, recommendation systems, etc. Motivated by concerns that user preferences/tastes can evolve over time, we consider the problem of _dueling bandits...
[ "non-stationary", "multi-armed bandits", "dueling bandits", "preference-based learning" ]
First optimal and adaptive dynamic regret upper bound in non-stationary dueling bandits under SST and STI, and proof of impossibility of same results outside of SST and STI.
15,236
2302.06595
title_snapshot
[ -0.03581445664167404, -0.019018299877643585, 0.008097860962152481, 0.034558188170194626, 0.029444769024848938, -0.006541188340634108, 0.009231945499777794, 0.026119768619537354, -0.0037312465719878674, -0.04128015413880348, -0.014843019656836987, 0.02320479042828083, -0.0450497604906559, -...
Self-supervised video pretraining yields robust and more human-aligned visual representations
https://openreview.net/forum?id=vF8ukt5l1R
[ "Nikhil Parthasarathy", "S. M. Ali Eslami", "Joao Carreira", "Olivier J Henaff" ]
Poster
null
Humans learn powerful representations of objects and scenes by observing how they evolve over time. Yet, outside of specific tasks that require explicit temporal understanding, static image pretraining remains the dominant paradigm for learning visual foundation models. We question this mismatch, and ask whether video ...
[ "self-supervised learning", "contrastive", "video pretraining", "representation learning", "visual representation", "human alignment", "robustness", "shape-bias", "saliency" ]
null
15,231
2210.06433
title_snapshot
[ 0.037358567118644714, 0.004030732437968254, 0.007280494552105665, 0.04150911793112755, 0.02727135829627514, 0.018065398558974266, 0.028684189543128014, 0.010631322860717773, -0.016369014978408813, -0.044211599975824356, -0.03151610866189003, -0.0055794003419578075, -0.07369405031204224, -0...
Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
https://openreview.net/forum?id=bzs4uPLXvi
[ "Miles Turpin", "Julian Michael", "Ethan Perez", "Samuel R. Bowman" ]
Poster
null
Large Language Models (LLMs) can achieve strong performance on many tasks by producing step-by-step reasoning before giving a final output, often referred to as chain-of-thought reasoning (CoT). It is tempting to interpret these CoT explanations as the LLM's process for solving a task. This level of transparency into L...
[ "Natural language processing", "large language models", "XAI", "explainability" ]
We show that chain-of-thought explanations can systematically misrepresent the real reasons influencing for model predictions.
15,229
2305.04388
title_snapshot
[ -0.022437313571572304, -0.008758739568293095, -0.03767386078834534, 0.06072961911559105, 0.02357422560453415, 0.005014271009713411, 0.0278935469686985, 0.0227794386446476, 0.007296910975128412, -0.006362466141581535, -0.0421268455684185, 0.04830997809767723, -0.06735467910766602, -0.005681...
f-Policy Gradients: A General Framework for Goal-Conditioned RL using f-Divergences
https://openreview.net/forum?id=EhhPtGsVAv
[ "Siddhant Agarwal", "Ishan Durugkar", "Peter Stone", "Amy Zhang" ]
Poster
null
Goal-Conditioned Reinforcement Learning (RL) problems often have access to sparse rewards where the agent receives a reward signal only when it has achieved the goal, making policy optimization a difficult problem. Several works augment this sparse reward with a learned dense reward function, but this can lead to sub...
[ "Goal Conditioned Reinforcement Learning", "Shaping Rewards", "Reward Design" ]
null
15,227
2310.06794
title_snapshot
[ -0.023751070722937584, -0.04681634157896042, 0.030913174152374268, 0.02609368972480297, 0.050386231392621994, 0.021880920976400375, -0.004667481407523155, -0.011803779751062393, -0.0370844230055809, -0.020030749961733818, -0.01329260878264904, 0.01309929322451353, -0.06845439225435257, -0....
Fine-Grained Human Feedback Gives Better Rewards for Language Model Training
https://openreview.net/forum?id=CSbGXyCswu
[ "Zeqiu Wu", "Yushi Hu", "Weijia Shi", "Nouha Dziri", "Alane Suhr", "Prithviraj Ammanabrolu", "Noah A. Smith", "Mari Ostendorf", "Hannaneh Hajishirzi" ]
Spotlight
null
Language models (LMs) often exhibit undesirable text generation behaviors, including generating false, toxic, or irrelevant outputs. Reinforcement learning from human feedback (RLHF)---where human preference judgments on LM outputs are transformed into a learning signal---has recently shown promise in addressing these...
[ "Language Model", "Reinforcement Learning with Human Feedback", "Long-Form Text Generation" ]
We introduce Fine-Grained RLHF, a framework that enables training and learning from fine-grained rewards: (1) density, providing a reward after each granular segment; and (2) mixing multiple reward models associated with different feedback types.
15,213
2306.01693
title_snapshot
[ -0.011284229345619678, 0.0038216752000153065, 0.01225153636187315, 0.045000530779361725, 0.06354562938213348, 0.00027425147709436715, 0.012487445026636124, 0.03920882195234299, -0.019976802170276642, -0.023147055879235268, -0.02005024626851082, 0.050860658288002014, -0.056526754051446915, ...
Disentangled Wasserstein Autoencoder for T-Cell Receptor Engineering
https://openreview.net/forum?id=Eb74zfBkWa
[ "Tianxiao Li", "Hongyu Guo", "Filippo Grazioli", "Mark Gerstein", "Martin Renqiang Min" ]
Poster
null
In protein biophysics, the separation between the functionally important residues (forming the active site or binding surface) and those that create the overall structure (the fold) is a well-established and fundamental concept. Identifying and modifying those functional sites is critical for protein engineering but co...
[ "protein engineering", "disentangled representation", "T cell receptor" ]
null
15,212
2210.08171
title_snapshot
[ 0.011896701529622078, -0.02214447408914566, -0.02027609385550022, 0.03945821523666382, 0.03889381140470505, -0.007518938276916742, 0.044987015426158905, -0.006083090323954821, 0.011228235438466072, -0.020385408774018288, -0.010218361392617226, 0.010081994347274303, -0.0868348479270935, 0.0...
A Diffusion-Model of Joint Interactive Navigation
https://openreview.net/forum?id=2yXExAl0FW
[ "Matthew Niedoba", "Jonathan Wilder Lavington", "Yunpeng Liu", "Vasileios Lioutas", "Justice Sefas", "Xiaoxuan Liang", "Dylan Green", "Setareh Dabiri", "Berend Zwartsenberg", "Adam Scibior", "Frank Wood" ]
Poster
null
Simulation of autonomous vehicle systems requires that simulated traffic participants exhibit diverse and realistic behaviors. The use of prerecorded real-world traffic scenarios in simulation ensures realism but the rarity of safety critical events makes large scale collection of driving scenarios expensive. In this p...
[ "Diffusion Models", "Trajecotry Forecasting", "Autonomous Vehicles", "Motion Forecasting", "Simulation" ]
We present a diffusion model which generates joint traffic scenes, and investigate its performance and test time conditioning methods.
15,178
2309.12508
title_snapshot
[ -0.02126654051244259, 0.004395178984850645, 0.00204316689632833, 0.044875312596559525, 0.05017969384789467, 0.022706015035510063, 0.03161117807030678, 0.016196051612496376, -0.012894753366708755, -0.07003890722990036, -0.0013911779969930649, -0.0018411979544907808, -0.06153037026524544, -0...
A Unifying Perspective on Multi-Calibration: Game Dynamics for Multi-Objective Learning
https://openreview.net/forum?id=ysqlhW0v26
[ "Nika Haghtalab", "Michael Jordan", "Eric Zhao" ]
Poster
null
We provide a unifying framework for the design and analysis of multi-calibrated predictors. By placing the multi-calibration problem in the general setting of multi-objective learning---where learning guarantees must hold simultaneously over a set of distributions and loss functions---we exploit connections to game dyn...
[ "multicalibration", "multi-objective learning", "learning theory", "calibration", "fairness", "games" ]
We use learning in game dynamics as a unifying framework for the design and analysis of multicalibrated predictors with guarantees improving on past results.
15,173
2302.10863
title_snapshot
[ -0.040680721402168274, -0.0019018909661099315, 0.008108258247375488, 0.019912585616111755, 0.02884860895574093, 0.04659344628453255, -0.00023162554134614766, -0.009575458243489265, -0.03440002724528313, -0.0514163039624691, 0.016074640676379204, 0.025509709492325783, -0.06776495277881622, ...
Data-driven Optimal Filtering for Linear Systems with Unknown Noise Covariances
https://openreview.net/forum?id=8S9Fbee743
[ "Shahriar Talebi", "Amirhossein Taghvaei", "Mehran Mesbahi" ]
Poster
null
This paper examines learning the optimal filtering policy, known as the Kalman gain, for a linear system with unknown noise covariance matrices using noisy output data. The learning problem is formulated as a stochastic policy optimiza- tion problem, aiming to minimize the output prediction error. This formulation prov...
[ "Optimal filtering", "data-driven control", "stochastic optimization", "learning" ]
null
15,169
2305.17836
title_snapshot
[ -0.0287658479064703, 0.01808416098356247, 0.0008082718122750521, 0.01690746657550335, 0.03348211571574211, 0.041978392750024796, 0.03703983500599861, 0.005544072948396206, 0.004398663528263569, -0.029813600704073906, -0.002042293781414628, 0.0049268812872469425, -0.09109535068273544, 0.001...
Hierarchical VAEs provide a normative account of motion processing in the primate brain
https://openreview.net/forum?id=1wOkHN9JK8
[ "Hadi Vafaii", "Jacob L. Yates", "Daniel A. Butts" ]
Poster
null
The relationship between perception and inference, as postulated by Helmholtz in the 19th century, is paralleled in modern machine learning by generative models like Variational Autoencoders (VAEs) and their hierarchical variants. Here, we evaluate the role of hierarchical inference and its alignment with brain functio...
[ "NeuroAI", "VAE", "Dorsal stream", "Hierarchical Bayesian Inference" ]
We used hierarchical VAEs and a novel simulation to test the hypothesis that brains perform hierarchical Bayesian inference about the state of the world.
15,168
null
null
[ -0.0012046745978295803, 0.05542287975549698, 0.02481011301279068, 0.00843378622084856, 0.015070374123752117, 0.03753774240612984, 0.06848491728305817, 0.014175300486385822, -0.06214800849556923, -0.03634846210479736, -0.01123072486370802, -0.005787793081253767, -0.06148071959614754, -0.014...
Optimal testing using combined test statistics across independent studies
https://openreview.net/forum?id=ZcuFDaMTYw
[ "Lasse Vuursteen", "Botond Szabo", "Aad van der Vaart", "Harry van Zanten" ]
Poster
null
Combining test statistics from independent trials or experiments is a popular method of meta-analysis. However, there is very limited theoretical understanding of the power of the combined test, especially in high-dimensional models considering composite hypotheses tests. We derive a mathematical framework to study sta...
[ "testing", "meta-analysis", "p-values", "e-values", "optimal", "combining trials" ]
We quantify the cost of compressing real valued test statistics such as p-values from $m$ independent trials concerning the same hypothesis into a single test.
15,167
2310.19541
title_snapshot
[ -0.011475017294287682, 0.011557876132428646, -0.00841535348445177, 0.04602827876806259, 0.0563398040831089, 0.022245053201913834, 0.06457583606243134, -0.021470729261636734, -0.0234844833612442, -0.04242924600839615, 0.024899108335375786, -0.0034027951769530773, -0.05568455532193184, -0.03...
Scale Alone Does not Improve Mechanistic Interpretability in Vision Models
https://openreview.net/forum?id=OZ7aImD4uQ
[ "Roland S. Zimmermann", "Thomas Klein", "Wieland Brendel" ]
Spotlight
null
In light of the recent widespread adoption of AI systems, understanding the internal information processing of neural networks has become increasingly critical. Most recently, machine vision has seen remarkable progress by scaling neural networks to unprecedented levels in dataset and model size. We here ask whether th...
[ "feature visualization", "interpretability", "explainability", "deep learning", "neural networks", "analysis", "activation maximization", "psychophysics" ]
We compare the mechanistic interpretability of vision models differing scale, architecture, training paradigm and dataset size and find that none of these design choices have any significant effect on the interpretability of individual units.
15,165
2307.05471
title_snapshot
[ -0.015922043472528458, 0.0052912067621946335, -0.0026304074563086033, 0.013808593153953552, 0.04112480953335762, 0.03682076185941696, 0.020041456446051598, 0.006192790810018778, -0.041090212762355804, -0.039101071655750275, -0.006670474540442228, 0.010033364407718182, -0.04718191921710968, ...
Tracking Most Significant Shifts in Nonparametric Contextual Bandits
https://openreview.net/forum?id=SGerL9HMrp
[ "Joe Suk", "Samory Kpotufe" ]
Poster
null
We study nonparametric contextual bandits where Lipschitz mean reward functions may change over time. We first establish the minimax dynamic regret rate in this less understood setting in terms of number of changes $L$ and total-variation $V$, both capturing all changes in distribution over context space, and argue tha...
[ "multi-armed bandits", "non-stationary", "contextual bandits", "nonparametric", "Lipschitz" ]
Optimal and adaptive dynamic regret bounds for non-stationary nonparametric contextual bandits in terms of new, tighter non-stationarity measure .
15,164
2307.05341
title_snapshot
[ -0.04650984704494476, -0.020627761259675026, -0.002501519164070487, 0.0405111163854599, 0.03862854838371277, 0.03253152593970299, 0.02146911434829235, 0.03786841034889221, -0.008534214459359646, -0.05433165282011032, -0.03324848785996437, 0.025844166055321693, -0.036292243748903275, -0.030...
SQ Lower Bounds for Non-Gaussian Component Analysis with Weaker Assumptions
https://openreview.net/forum?id=Xp68yXQiRk
[ "Ilias Diakonikolas", "Daniel Kane", "Lisheng Ren", "Yuxin Sun" ]
Poster
null
We study the complexity of Non-Gaussian Component Analysis (NGCA) in the Statistical Query (SQ) model. Prior work developed a methodology to prove SQ lower bounds for NGCA that have been applicable to a wide range of contexts. In particular, it was known that for any univariate distribution $A$ satisfying certain condi...
[ "Non-Gaussian Component Analysis" ]
null
15,157
2403.04744
title_snapshot
[ -0.013609792105853558, -0.016837161034345627, 0.019228773191571236, 0.029113538563251495, 0.015381383709609509, 0.05000367760658264, 0.027440672740340233, -0.007419769186526537, -0.04243229702115059, -0.02657848596572876, -0.030546963214874268, -0.019363125786185265, -0.06914415955543518, ...
Precise asymptotic generalization for multiclass classification with overparameterized linear models
https://openreview.net/forum?id=cRGINXQWem
[ "David Xing Wu", "Anant Sahai" ]
Spotlight
null
We study the asymptotic generalization of an overparameterized linear model for multiclass classification under the Gaussian covariates bi-level model introduced in Subramanian et al. (NeurIPS'22), where the number of data points, features, and classes all grow together. We fully resolve the conjecture posed in Subrama...
[ "overparameterized", "multiclass", "classification", "theory", "generalization", "interpolation", "bi-level", "Gaussian model" ]
null
15,155
2306.13255
title_snapshot
[ -0.00878884457051754, -0.01927117258310318, -0.010623696260154247, 0.008827480487525463, 0.01782943867146969, 0.03080666810274124, 0.0372464656829834, -0.03452553600072861, -0.02831103466451168, -0.02525363303720951, -0.006703937891870737, -0.00975827593356371, -0.11008304357528687, 0.0064...
Fair Adaptive Experiments
https://openreview.net/forum?id=PMvudWa53L
[ "Waverly Wei", "Xinwei Ma", "Jingshen Wang" ]
Poster
null
Randomized experiments have been the gold standard for assessing the effectiveness of a treatment, policy, or intervention, spanning various fields, including social sciences, biomedical studies, and e-commerce. The classical complete randomization approach assigns treatments based on a pre-specified probability and ma...
[ "Adaptive Randomized Experiment; Adaptive Design; Causal Inference" ]
null
15,151
2310.16290
title_snapshot
[ -0.009972257539629936, -0.035837993025779724, -0.027670662850141525, 0.030620049685239792, 0.031763944774866104, 0.027042772620916367, 0.04352251812815666, 0.007742759771645069, -0.03895845636725426, -0.047838520258665085, 0.0002297228784300387, 0.00185685814358294, -0.059942834079265594, ...
Diverse Shape Completion via Style Modulated Generative Adversarial Networks
https://openreview.net/forum?id=yVMlYSL1Bp
[ "Wesley Khademi", "Li Fuxin" ]
Poster
null
Shape completion aims to recover the full 3D geometry of an object from a partial observation. This problem is inherently multi-modal since there can be many ways to plausibly complete the missing regions of a shape. Such diversity would be indicative of the underlying uncertainty of the shape and could be preferable f...
[ "multimodal shape completion", "point cloud completion", "3d shape generation", "generative modeling", "generative adversarial networks" ]
null
15,143
2311.11184
title_snapshot
[ 0.021115245297551155, -0.040846142917871475, -0.01819307543337345, 0.07419014722108841, 0.015588157810270786, 0.050289370119571686, -0.004374403972178698, 0.00814568717032671, -0.05241981893777847, -0.0989324077963829, -0.06607179343700409, -0.0029040430672466755, -0.05264474079012871, -0....
UNSSOR: Unsupervised Neural Speech Separation by Leveraging Over-determined Training Mixtures
https://openreview.net/forum?id=T5h69frFF7
[ "Zhong-Qiu Wang", "Shinji Watanabe" ]
Poster
null
In reverberant conditions with multiple concurrent speakers, each microphone acquires a mixture signal of multiple speakers at a different location. In over-determined conditions where the microphones out-number speakers, we can narrow down the solutions to speaker images and realize unsupervised speech separation by l...
[ "Speech separation", "microphone array processing", "deep learning" ]
We show that DNNs can be trained directly on multi-speaker mixtures to realize unsupervised neural speech separation if the training mixtures are over-determined (i.e., more microphones than speakers).
15,142
2305.20054
title_snapshot
[ -0.004342186264693737, -0.019773824140429497, -0.0002444240089971572, 0.0022789733484387398, 0.030205123126506805, 0.03397731855511665, 0.029927441850304604, 0.0066549391485750675, -0.05089952051639557, -0.04503302276134491, -0.013330161571502686, 0.05633660778403282, -0.034503281116485596, ...
Understanding the detrimental class-level effects of data augmentation
https://openreview.net/forum?id=yageaKlk7S
[ "Polina Kirichenko", "Mark Ibrahim", "Randall Balestriero", "Diane Bouchacourt", "Shanmukha Ramakrishna Vedantam", "Hamed Firooz", "Andrew Gordon Wilson" ]
Poster
null
Data augmentation (DA) encodes invariance and provides implicit regularization critical to a model's performance in image classification tasks. However, while DA improves average accuracy, recent studies have shown that its impact can be highly class dependent: achieving optimal average accuracy comes at the cost of si...
[ "data augmentation", "class-dependent bias" ]
null
15,140
2401.01764
title_snapshot
[ 0.007452302612364292, -0.012514494359493256, -0.030585668981075287, 0.07929183542728424, 0.02807973325252533, -0.013931302353739738, 0.03185635060071945, -0.01524271722882986, -0.03247442841529846, -0.044123437255620956, -0.03806139901280403, -0.004373631440103054, -0.07020852714776993, -0...
Versatile Energy-Based Probabilistic Models for High Energy Physics
https://openreview.net/forum?id=j0U6XJubbP
[ "Taoli Cheng", "Aaron Courville" ]
Poster
null
As a classical generative modeling approach, energy-based models have the natural advantage of flexibility in the form of the energy function. Recently, energy-based models have achieved great success in modeling high-dimensional data in computer vision and natural language processing. In line with these advancements, ...
[ "Generative modeling", "Energy-based models", "Out-of-distribution detection", "Sciences", "Application", "Physics" ]
null
15,125
2302.00695
title_snapshot
[ 0.0028269251342862844, 0.01485417690128088, -0.028072906658053398, 0.035371892154216766, 0.01844249665737152, -0.0009090199018828571, -0.002764882752671838, 0.02356070838868618, -0.05093101039528847, -0.028780685737729073, -0.015506966039538383, -0.003010221291333437, -0.054028917104005814, ...
Compositional Generalization from First Principles
https://openreview.net/forum?id=LqOQ1uJmSx
[ "Thaddäus Wiedemer", "Prasanna Mayilvahanan", "Matthias Bethge", "Wieland Brendel" ]
Poster
null
Leveraging the compositional nature of our world to expedite learning and facilitate generalization is a hallmark of human perception. In machine learning, on the other hand, achieving compositional generalization has proven to be an elusive goal, even for models with explicit compositional priors. To get a better hand...
[ "compositional generalization", "compositionality", "generalization", "combinatorial generalization", "out-of-distribution", "out-of-domain", "identifiability", "disentanglement", "object-centric learning", "DSprites" ]
We derive sufficient conditions on the data generating process and model architecture under which compositional generalization provably occurs.
15,124
2307.05596
title_snapshot
[ -0.00032058110809884965, -0.0055638146586716175, 0.0035184365697205067, 0.05420537292957306, 0.042135000228881836, 0.01662868820130825, 0.019813820719718933, 0.014979910105466843, -0.014902258291840553, -0.02686990797519684, -0.033859334886074066, 0.0014906810829415917, -0.06784088909626007,...
NAS-X: Neural Adaptive Smoothing via Twisting
https://openreview.net/forum?id=A9mHph8GJk
[ "Dieterich Lawson", "Michael Y. Li", "Scott Linderman" ]
Poster
null
Sequential latent variable models (SLVMs) are essential tools in statistics and machine learning, with applications ranging from healthcare to neuroscience. As their flexibility increases, analytic inference and model learning can become challenging, necessitating approximate methods. Here we introduce neural adaptive ...
[ "sequence models", "probabilistic inference", "reweighted wake-sleep", "sequential monte carlo", "smoothing", "mechanistic models" ]
We introduce a method for fitting sequential latent variable models that combines the benefits of reweighted wake-sleep and smoothing sequential Monte Carlo.
15,111
2308.14864
title_snapshot
[ -0.030856797471642494, -0.032899580895900726, -0.0050905440002679825, 0.011876329779624939, 0.03257365524768829, 0.04682319238781929, 0.04262194037437439, -0.0038190893828868866, -0.0468277670443058, -0.04618340730667114, -0.006005145609378815, 0.002072516828775406, -0.038738884031772614, ...
SpecTr: Fast Speculative Decoding via Optimal Transport
https://openreview.net/forum?id=SdYHLTCC5J
[ "Ziteng Sun", "Ananda Theertha Suresh", "Jae Hun Ro", "Ahmad Beirami", "Himanshu Jain", "Felix Yu" ]
Poster
null
Autoregressive sampling from large language models has led to state-of-the-art results in several natural language tasks. However, autoregressive sampling generates tokens one at a time making it slow, and even prohibitive in certain tasks. One way to speed up sampling is *speculative decoding*: use a small model to sa...
[ "autoregressive sampling; computation efficiency; optimal transport" ]
null
15,090
2310.15141
title_snapshot
[ -0.020077310502529144, -0.017636248841881752, -0.011266034096479416, 0.05787607282400131, 0.03408503159880638, 0.04301255941390991, 0.013816360384225845, 0.01606043055653572, -0.017121633514761925, -0.05174512788653374, -0.010664969682693481, 0.013557713478803635, -0.08069043606519699, -0....
Fair, Polylog-Approximate Low-Cost Hierarchical Clustering
https://openreview.net/forum?id=cAPMmCl2f3
[ "Marina Knittel", "Max Springer", "John P Dickerson", "MohammadTaghi Hajiaghayi" ]
Poster
null
Research in fair machine learning, and particularly clustering, has been crucial in recent years given the many ethical controversies that modern intelligent systems have posed. Ahmadian et al. [2020] established the study of fairness in hierarchical clustering, a stronger, more structured variant of its well-known fla...
[ "Fair machine learning", "hierarchical clustering", "clustering" ]
We propose the first polylog-approximate fair hierarchical clustering algorithm optimizing for Dasgupta's cost function.
15,086
2311.12501
title_snapshot
[ -0.01263806689530611, -0.0007716971449553967, -0.010243374854326248, 0.024615639820694923, 0.04537283256649971, 0.017207512632012367, -0.004885273985564709, -0.013310201466083527, -0.041441500186920166, -0.03162338584661484, 0.013966798782348633, -0.021961690858006477, -0.06624777615070343, ...
Minimax-Optimal Location Estimation
https://openreview.net/forum?id=JeKXmYb4kd
[ "Shivam Gupta", "Jasper C.H. Lee", "Eric Price", "Paul Valiant" ]
Poster
null
Location estimation is one of the most basic questions in parametric statistics. Suppose we have a known distribution density $f$, and we get $n$ i.i.d. samples from $f(x-\mu)$ for some unknown shift $\mu$. The task is to estimate $\mu$ to high accuracy with high probability. The maximum likelihood estimator (MLE) is ...
[ "location estimation", "minimax estimation" ]
null
15,079
null
null
[ -0.021031152456998825, 0.0057634031400084496, 0.00955063197761774, 0.013576236553490162, 0.05276121199131012, 0.05107039958238602, 0.03146626800298691, 0.03440477326512337, -0.042946986854076385, -0.047330956906080246, 0.010106916539371014, -0.025059470906853676, -0.0523819625377655, 0.017...
Posterior Sampling with Delayed Feedback for Reinforcement Learning with Linear Function Approximation
https://openreview.net/forum?id=RiyH3z7oIF
[ "Nikki Lijing Kuang", "Ming Yin", "Mengdi Wang", "Yu-Xiang Wang", "Yian Ma" ]
Poster
null
Recent studies in reinforcement learning (RL) have made significant progress by leveraging function approximation to alleviate the sample complexity hurdle for better performance. Despite the success, existing provably efficient algorithms typically rely on the accessibility of immediate feedback upon taking actions. T...
[ "Posterior Sampling", "Reinforcement Learning Theory", "Linear Markov Decision Processes", "Delayed Feedback", "Langevin Monte Carlo" ]
We provide the first analysis for the class of PS algorithms to handle delayed feedback in RL frameworks.
15,077
2310.18919
title_snapshot
[ -0.01854972541332245, -0.016204126179218292, -0.016028309240937233, 0.052761420607566833, 0.041978005319833755, 0.05857565253973007, 0.012728232890367508, -0.011700094677507877, -0.03864801675081253, -0.05876482278108597, 0.008290303871035576, 0.0158500038087368, -0.05430148169398308, -0.0...
Projection-Free Methods for Solving Nonconvex-Concave Saddle Point Problems
https://openreview.net/forum?id=WO1kHC5Lfz
[ "Morteza Boroun", "Erfan Yazdandoost Hamedani", "Afrooz Jalilzadeh" ]
Poster
null
In this paper, we investigate a class of constrained saddle point (SP) problems where the objective function is nonconvex-concave and smooth. This class of problems has wide applicability in machine learning, including robust multi-class classification and dictionary learning. Several projection-based primal-dual metho...
[ "Saddle Point Problem", "Projection-free method" ]
null
15,064
2306.11944
title_snapshot
[ -0.023831387981772423, -0.0179994348436594, -0.0012370811309665442, 0.04212145507335663, 0.031226856634020805, 0.049890194088220596, 0.0051749818958342075, -0.01525223907083273, -0.026568468660116196, -0.053973857313394547, -0.01674540527164936, 0.001760188490152359, -0.06442203372716904, ...
A polar prediction model for learning to represent visual transformations
https://openreview.net/forum?id=hyPUZX03Ks
[ "Pierre-Etienne H Fiquet", "Eero P Simoncelli" ]
Poster
null
All organisms make temporal predictions, and their evolutionary fitness level depends on the accuracy of these predictions. In the context of visual perception, the motions of both the observer and objects in the scene structure the dynamics of sensory signals, allowing for partial prediction of future signals based on...
[ "video prediction", "neural coding", "symmetry discovery", "self-supervised representation-learning" ]
null
15,063
2303.03432
title_snapshot
[ 0.0024969724472612143, 0.011969211511313915, 0.03127416595816612, 0.01904882863163948, 0.029598621651530266, 0.027949579060077667, 0.03293178975582123, 0.020011214539408684, -0.053036052733659744, -0.029065018519759178, -0.033126965165138245, -0.01997828669846058, -0.06774440407752991, -0....
Kullback-Leibler Maillard Sampling for Multi-armed Bandits with Bounded Rewards
https://openreview.net/forum?id=xF89MjFbWp
[ "Hao Qin", "Kwang-Sung Jun", "Chicheng Zhang" ]
Poster
null
We study $K$-armed bandit problems where the reward distributions of the arms are all supported on the $[0,1]$ interval. Maillard sampling\cite{maillard13apprentissage}, an attractive alternative to Thompson sampling, has recently been shown to achieve competitive regret guarantees in the sub-Gaussian reward setting\ci...
[ "multi-armed bandits", "bounded rewards" ]
null
15,060
2304.14989
title_snapshot
[ -0.04763812944293022, 0.004665798041969538, 0.00910720694810152, 0.02361835166811943, 0.04660634696483612, 0.00494099548086524, 0.04800931736826897, 0.005541646387428045, -0.02282867208123207, -0.05035467445850372, -0.015175148844718933, 0.004767242819070816, -0.034720856696367264, -0.0263...
No Train No Gain: Revisiting Efficient Training Algorithms For Transformer-based Language Models
https://openreview.net/forum?id=thbXgJ8gNK
[ "Jean Kaddour", "Oscar Key", "Piotr Nawrot", "Pasquale Minervini", "Matt Kusner" ]
Poster
null
The computation necessary for training Transformer-based language models has skyrocketed in recent years. This trend has motivated research on efficient training algorithms designed to improve training, validation, and downstream performance faster than standard training. In this work, we revisit three categories of su...
[ "language models", "transformers", "efficient training" ]
We investigate efficient training algorithms for Transformer Language models and find many situations in which they have only marginal improvements over standard training.
15,048
2307.06440
title_snapshot
[ -0.028483236208558083, -0.03054870292544365, -0.02072075568139553, 0.036975566297769547, 0.013040016405284405, 0.034783393144607544, 0.042925167828798294, 0.022456126287579536, -0.022426914423704147, -0.009983151219785213, -0.03672827035188675, 0.0691927894949913, -0.06195962056517601, 0.0...
Modelling Cellular Perturbations with the Sparse Additive Mechanism Shift Variational Autoencoder
https://openreview.net/forum?id=DzaCE00jGV
[ "Michael Bereket", "Theofanis Karaletsos" ]
Poster
null
Generative models of observations under interventions have been a vibrant topic of interest across machine learning and the sciences in recent years. For example, in drug discovery, there is a need to model the effects of diverse interventions on cells in order to characterize unknown biological mechanisms of action. W...
[ "Disentagled representation learning", "VAE", "generative models", "sparse mechanism shift", "perturbation modeling", "cellular modeling" ]
We propose a new VAE structure, SAMS-VAE, which captures explicit perturbation latent variables and sparsity effects to model sparse mechanism shift in latent space for applications such as compositional perturbation modeling of cells.
15,039
2311.02794
title_snapshot
[ 0.016118193045258522, -0.022942958399653435, -0.03667670860886574, 0.022713730111718178, 0.036529961973428726, 0.015642676502466202, 0.06485813111066818, -0.025186948478221893, -0.051953788846731186, -0.044365573674440384, 0.012904606759548187, 0.0008432787144556642, -0.07189198583364487, ...
ParaFuzz: An Interpretability-Driven Technique for Detecting Poisoned Samples in NLP
https://openreview.net/forum?id=DD0QJvPbTD
[ "Lu Yan", "ZHUO ZHANG", "Guanhong Tao", "Kaiyuan Zhang", "Xuan Chen", "Guangyu Shen", "Xiangyu Zhang" ]
Poster
null
Backdoor attacks have emerged as a prominent threat to natural language processing (NLP) models, where the presence of specific triggers in the input can lead poisoned models to misclassify these inputs to predetermined target classes. Current detection mechanisms are limited by their inability to address more covert b...
[ "NLP", "backdoor attack", "fuzzing" ]
we propose an innovative test-time poisoned sample detection framework that hinges on the interpretability of model predictions, grounded in the semantic meaning of inputs.
15,028
2308.02122
title_snapshot
[ -0.015263351611793041, -0.039497893303632736, -0.03571251407265663, 0.04089672118425369, 0.045875370502471924, 0.013341514393687248, 0.06340125948190689, 0.013286576606333256, 0.005400849971920252, -0.013586451299488544, -0.027625611051917076, 0.03427315875887871, -0.05880989879369736, -0....
HiNeRV: Video Compression with Hierarchical Encoding-based Neural Representation
https://openreview.net/forum?id=CpoS56pYnU
[ "Ho Man Kwan", "Ge Gao", "Fan Zhang", "Andy Gower", "David Bull" ]
Poster
null
Learning-based video compression is currently a popular research topic, offering the potential to compete with conventional standard video codecs. In this context, Implicit Neural Representations (INRs) have previously been used to represent and compress image and video content, demonstrating relatively high decoding s...
[ "Video compression", "Implicit neural representations" ]
null
15,020
2306.09818
title_snapshot
[ -0.012551981024444103, -0.012729734182357788, -0.018390534445643425, 0.018059706315398216, 0.027466682717204094, 0.030585814267396927, -0.0034516158048063517, -0.015286918729543686, -0.033299777656793594, -0.05365287512540817, 0.007006203755736351, -0.02798675000667572, -0.036418113857507706...
No-Regret Online Prediction with Strategic Experts
https://openreview.net/forum?id=AesN5bYnJr
[ "Omid Sadeghi", "Maryam Fazel" ]
Poster
null
We study a generalization of the online binary prediction with expert advice framework where at each round, the learner is allowed to pick $m\geq 1$ experts from a pool of $K$ experts and the overall utility is a modular or submodular function of the chosen experts. We focus on the setting in which experts act strategi...
[ "incentive-compatible", "online prediction with expert advice", "forecasting" ]
We propose the first no-regret and incentive-compatible algorithms for a generalization of the experts problem (called the "m-experts" problem)
15,012
2305.15331
title_snapshot
[ -0.044149890542030334, -0.03658801317214966, 0.013509114272892475, 0.03127589821815491, 0.04253807291388512, 0.026542769744992256, 0.013513552024960518, 0.021152207627892494, -0.028293147683143616, -0.03835131227970123, -0.02834750898182392, 0.03534819930791855, -0.06637784093618393, -0.04...
Uncovering motifs of concurrent signaling across multiple neuronal populations
https://openreview.net/forum?id=u39QQh5L8Q
[ "Evren Gokcen", "Anna Ivic Jasper", "Alison Xu", "Adam Kohn", "Christian K. Machens", "Byron M. Yu" ]
Spotlight
null
Modern recording techniques now allow us to record from distinct neuronal populations in different brain networks. However, especially as we consider multiple (more than two) populations, new conceptual and statistical frameworks are needed to characterize the multi-dimensional, concurrent flow of signals among these p...
[ "neuroscience", "multi-population neural recordings", "dimensionality reduction", "latent variable models", "Gaussian processes" ]
We developed a dimensionality reduction framework for characterizing the multi-dimensional, concurrent flow of signals across multiple neuronal populations.
15,007
null
null
[ -0.027490681037306786, -0.014005773700773716, -0.014107913710176945, 0.02161444164812565, 0.018662063404917717, 0.04455455765128136, 0.04719216004014015, 0.019190100952982903, -0.062007129192352295, -0.067147396504879, 0.016344016417860985, -0.024419430643320084, -0.07387517392635345, 0.00...
ELDEN: Exploration via Local Dependencies
https://openreview.net/forum?id=sL4pJBXkxu
[ "Zizhao Wang", "Jiaheng Hu", "Peter Stone", "Roberto Martín-Martín" ]
Poster
null
Tasks with large state space and sparse rewards present a longstanding challenge to reinforcement learning. In these tasks, an agent needs to explore the state space efficiently until it finds a reward. To deal with this problem, the community has proposed to augment the reward function with intrinsic reward, a bonus s...
[ "reinforcement learning; intrinsic motivation; exploration" ]
An exploration method for reinforcement learning based on novel local dependencies.
14,993
2310.08702
title_snapshot
[ -0.04853261634707451, 0.005701275076717138, -0.016170969232916832, 0.07232844829559326, 0.058009762316942215, 0.008606630377471447, 0.0017655184492468834, -0.03185628727078438, -0.022806813940405846, -0.03491191938519478, -0.03183164820075035, -0.00791023951023817, -0.05618561804294586, -0...
Improved Algorithms for Stochastic Linear Bandits Using Tail Bounds for Martingale Mixtures
https://openreview.net/forum?id=TXoZiUZywf
[ "Hamish Flynn", "David Reeb", "Melih Kandemir", "Jan Peters" ]
Oral
null
We present improved algorithms with worst-case regret guarantees for the stochastic linear bandit problem. The widely used "optimism in the face of uncertainty" principle reduces a stochastic bandit problem to the construction of a confidence sequence for the unknown reward function. The performance of the resulting ba...
[ "Linear bandits", "confidence sequences", "martingales", "convex optimization", "cumulative regret", "regret analysis" ]
Based on novel mixture martingales, we obtain tighter confidence bounds for linear bandits resulting in better algorithms with performance guarantees.
14,987
2309.14298
title_snapshot
[ -0.002921555656939745, 0.0033655797597020864, 0.00033324735704809427, 0.04474622756242752, 0.04116854444146156, 0.02187725156545639, 0.04382593557238579, 0.007922424003481865, -0.01514066755771637, -0.04802023991942406, -0.021541181951761246, 0.00897915568202734, -0.06103324517607689, -0.0...
How to Scale Your EMA
https://openreview.net/forum?id=DkeeXVdQyu
[ "Dan Busbridge", "Jason Ramapuram", "Pierre Ablin", "Tatiana Likhomanenko", "Eeshan Gunesh Dhekane", "Xavier Suau", "Russell Webb" ]
Spotlight
null
Preserving training dynamics across batch sizes is an important tool for practical machine learning as it enables the trade-off between batch size and wall-clock time. This trade-off is typically enabled by a scaling rule, for example, in stochastic gradient descent, one should scale the learning rate linearly with the...
[ "Optimization", "scaling rules", "EMA", "exponential moving average", "self-supervised learning", "pseudo-labelling", "semi-supervised learning", "BYOL", "distillation", "speech", "vision" ]
We derive and verify a scaling law for model exponential moving averages, allowing the systematic scaling of model classes depending on EMA, like distillation self-supervised methods.
14,973
2307.13813
title_snapshot
[ -0.018690133467316628, -0.021513886749744415, -0.004857162944972515, -0.01133787352591753, 0.04924057051539421, 0.030077604576945305, 0.04591402783989906, -0.017772983759641647, -0.06012285500764847, -0.01503976620733738, 0.003977854736149311, -0.012113172560930252, -0.05110783129930496, 0...
$H$-Consistency Bounds: Characterization and Extensions
https://openreview.net/forum?id=nI7EmXq2PL
[ "Anqi Mao", "Mehryar Mohri", "Yutao Zhong" ]
Poster
null
A series of recent publications by Awasthi et al. have introduced the key notion of *$H$-consistency bounds* for surrogate loss functions. These are upper bounds on the zero-one estimation error of any predictor in a hypothesis set, expressed in terms of its surrogate loss estimation error. They are both non-asymptoti...
[ "consistency", "H-consistency", "characterization", "learning theory" ]
null
14,971
null
null
[ -0.024769801646471024, 0.00870975386351347, -0.032311443239450455, 0.036174602806568146, 0.03352503851056099, 0.0407290980219841, 0.02790277823805809, -0.03341548889875412, -0.029491588473320007, -0.0386754609644413, -0.0037222751416265965, 0.0023550335317850113, -0.07582388818264008, -0.0...
Single-Pass Pivot Algorithm for Correlation Clustering. Keep it simple!
https://openreview.net/forum?id=lkEiOZlmPm
[ "Konstantin Makarychev", "Sayak Chakrabarty" ]
Poster
null
We show that a simple single-pass semi-streaming variant of the Pivot algorithm for Correlation Clustering gives a (3+eps)-approximation using O(n/eps) words of memory. This is a slight improvement over the recent results of Cambus, Kuhn, Lindy, Pai, and Uitto, who gave a (3+eps)-approximation using O(n log n) words of...
[ "correlation clustering", "Pivot algorithm", "streaming" ]
We show that a variant of the Pivot algorithm for Correlation Clustering gives a (3+eps) approximation in the single-pass semi-streaming model with O(n/eps) words of memory.
14,957
2305.13560
title_snapshot
[ -0.0067841606214642525, -0.0034993079025298357, 0.001383759779855609, 0.04568777605891228, 0.03497225418686867, 0.055977270007133484, 0.009286243468523026, 0.017911089584231377, -0.030606292188167572, -0.031604647636413574, -0.005706248804926872, -0.02095157280564308, -0.0679410919547081, ...
Model-free Posterior Sampling via Learning Rate Randomization
https://openreview.net/forum?id=IrjXmIKFyx
[ "Daniil Tiapkin", "Denis Belomestny", "Daniele Calandriello", "Eric Moulines", "Remi Munos", "Alexey Naumov", "Pierre Perrault", "Michal Valko", "Pierre MENARD" ]
Poster
null
In this paper, we introduce Randomized Q-learning (RandQL), a novel randomized model-free algorithm for regret minimization in episodic Markov Decision Processes (MDPs). To the best of our knowledge, RandQL is the first tractable model-free posterior sampling-based algorithm. We analyze the performance of RandQL in bot...
[ "reinforcement learning", "exploration", "q-learning" ]
learning rate randomization in Q-learning gives provably efficient exploration
14,956
2310.18186
title_snapshot
[ -0.026801468804478645, -0.014216414652764797, -0.008947783149778843, 0.06108352914452553, 0.03989782929420471, 0.028618067502975464, 0.0023855126928538084, -0.004466341808438301, -0.018752263858914375, -0.05025031790137291, -0.02070481702685356, 0.015116252936422825, -0.06118163466453552, ...
A Unified, Scalable Framework for Neural Population Decoding
https://openreview.net/forum?id=sw2Y0sirtM
[ "Mehdi Azabou", "Vinam Arora", "Venkataramana Ganesh", "Ximeng Mao", "Santosh B Nachimuthu", "Michael Jacob Mendelson", "Blake Aaron Richards", "Matthew G Perich", "Guillaume Lajoie", "Eva L Dyer" ]
Poster
null
Our ability to use deep learning approaches to decipher neural activity would likely benefit from greater scale, in terms of both the model size and the datasets. However, the integration of many neural recordings into one unified model is challenging, as each recording contains the activity of different neurons from d...
[ "neural population", "brain decoder", "transformer", "tokenization", "sequence-to-sequence", "electrophysiology", "brain-computer interfaces" ]
This paper introduces a scalable and unified learning framework for efficient, large-scale neural decoding across diverse multi-lab multi-animal multi-session neural recordings.
14,941
2310.16046
title_snapshot
[ -0.0001449637347832322, -0.02580822817981243, -0.01236194372177124, 0.026331085711717606, 0.021920347586274147, 0.05570492520928383, 0.03484316170215607, 0.03812440112233162, -0.05316593125462532, -0.029168792068958282, 0.0003361297131050378, -0.007608085870742798, -0.06114181876182556, -0...
A Trichotomy for Transductive Online Learning
https://openreview.net/forum?id=iSd8g75QvP
[ "Steve Hanneke", "Shay Moran", "Jonathan Shafer" ]
Poster
null
We present new upper and lower bounds on the number of learner mistakes in the `transductive' online learning setting of Ben-David, Kushilevitz and Mansour (1997). This setting is similar to standard online learning, except that the adversary fixes a sequence of instances $x_1,\dots,x_n$ to be labeled at the start ...
[ "Online Learning", "Transductive Online Learning", "Offline Learning", "Mistake Bound" ]
A qualitative trichotomy and some quantitative mistakes bounds relating the transductive online learning setting to known combinatorial dimensions.
14,914
2311.06428
title_snapshot
[ -0.027755776420235634, 0.0001802025071810931, -0.02586582489311695, 0.03383904695510864, 0.0270015187561512, 0.014014547690749168, 0.028134029358625412, 0.01587584987282753, -0.022543709725141525, -0.012067911215126514, 0.006944514345377684, 0.020246675238013268, -0.06965865939855576, -0.0...
Towards Automated Circuit Discovery for Mechanistic Interpretability
https://openreview.net/forum?id=89ia77nZ8u
[ "Arthur Conmy", "Augustine N. Mavor-Parker", "Aengus Lynch", "Stefan Heimersheim", "Adrià Garriga-Alonso" ]
Spotlight
null
Through considerable effort and intuition, several recent works have reverse-engineered nontrivial behaviors of transformer models. This paper systematizes the mechanistic interpretability process they followed. First, researchers choose a metric and dataset that elicit the desired model behavior. Then, they apply acti...
[ "Mechanistic Interpretability", "Pruning", "Science of Deep Learning", "AI Safety" ]
We identify the common workflow for mechanistic interpretability work, and automate its “systematic ablations” step with a new algorithm, ACDC.
14,912
2304.14997
title_snapshot
[ -0.01093136053532362, -0.02052059955894947, -0.03328641131520271, 0.04766807705163956, 0.03338523581624031, 0.033501867204904556, 0.007963428273797035, -0.0031705074943602085, -0.009318198077380657, -0.005860756617039442, 0.023159639909863472, 0.02330057881772518, -0.040353674441576004, 0....
Generating Behaviorally Diverse Policies with Latent Diffusion Models
https://openreview.net/forum?id=nafgeYknRT
[ "Shashank Hegde", "Sumeet Batra", "K.R. Zentner", "Gaurav S. Sukhatme" ]
Poster
null
Recent progress in Quality Diversity Reinforcement Learning (QD-RL) has enabled learning a collection of behaviorally diverse, high performing policies. However, these methods typically involve storing thousands of policies, which results in high space-complexity and poor scaling to additional behaviors. Condensing the...
[ "Latent Diffusion", "Quality Diversity", "Reinforcement Learning", "Graph Neural Networks" ]
A latent diffusion model that generates a behaviorally diverse set of neural network policies for robotic control
14,894
2305.18738
title_snapshot
[ -0.01746223494410515, -0.047148797661066055, -0.04902886226773262, 0.08239630609750748, 0.05167955532670021, 0.03745856136083603, 0.0049120099283754826, -0.023522110655903816, -0.030409259721636772, -0.041653383523225784, -0.005181537475436926, -0.013193751685321331, -0.056319091469049454, ...
Distributed Personalized Empirical Risk Minimization
https://openreview.net/forum?id=KoQgA0coZ9
[ "Yuyang Deng", "Mohammad Mahdi Kamani", "Pouria Mahdavinia", "Mehrdad Mahdavi" ]
Poster
null
This paper advocates a new paradigm Personalized Empirical Risk Minimization (PERM) to facilitate learning from heterogeneous data sources without imposing stringent constraints on computational resources shared by participating devices. In PERM, we aim at learning a distinct model for each client by personalizing ...
[ "distributed learning", "heterogeneous data", "heterogeneous system", "convergence analysis" ]
null
14,891
2310.17761
title_snapshot
[ -0.010375364683568478, -0.01330509688705206, 0.005090765189379454, 0.038364704698324203, 0.05733194947242737, 0.03837772458791733, 0.018665991723537445, -0.03580586239695549, -0.026673974469304085, -0.04722512140870094, -0.00015340362733695656, -0.002462243428453803, -0.04769422858953476, ...
Structured Prediction with Stronger Consistency Guarantees
https://openreview.net/forum?id=YZ7ip645Ra
[ "Anqi Mao", "Mehryar Mohri", "Yutao Zhong" ]
Poster
null
We present an extensive study of surrogate losses for structured prediction supported by *$H$-consistency bounds*. These are recently introduced guarantees that are more relevant to learning than Bayes-consistency, since they are not asymptotic and since they take into account the hypothesis set $H$ used. We first show...
[ "structured prediction", "consistency", "learning theory", "natural language processing" ]
null
14,890
null
null
[ -0.001948836725205183, -0.006825065240263939, -0.010267309844493866, 0.05166582763195038, 0.04220600426197052, 0.03161689266562462, 0.020517153665423393, -0.027437172830104828, -0.008869728073477745, -0.029402438551187515, -0.0029028577264398336, 0.014901041984558105, -0.08516321331262589, ...
Feature Learning for Interpretable, Performant Decision Trees
https://openreview.net/forum?id=PYEgC56flW
[ "Jack Henry Good", "Torin Kovach", "Kyle Miller", "Artur Dubrawski" ]
Poster
null
Decision trees are regarded for high interpretability arising from their hierarchical partitioning structure built on simple decision rules. However, in practice, this is not realized because axis-aligned partitioning of realistic data results in deep trees, and because ensemble methods are used to mitigate overfitting...
[ "explainability", "interpretability", "decision tree", "feature learning" ]
null
14,889
null
null
[ -0.018270110711455345, -0.006407521199434996, -0.007186467293649912, 0.03115415945649147, 0.03875483199954033, 0.042352814227342606, 0.04025682806968689, -0.03297893702983856, -0.035258177667856216, -0.026771564036607742, -0.011898127384483814, -0.004985433537513018, -0.08852378278970718, ...
Advice Querying under Budget Constraint for Online Algorithms
https://openreview.net/forum?id=QpZubU4yD9
[ "Ziyad Benomar", "Vianney Perchet" ]
Poster
null
Several problems have been extensively studied in the learning-augmented setting, where the algorithm has access to some, possibly incorrect, predictions. However, it is assumed in most works that the predictions are provided to the algorithm as input, with no constraint on their size. In this paper, we consider algori...
[ "online algorithms", "competitive ratio", "learning augmented algorithms", "scheduling", "ski-rental", "secretary" ]
null
14,864
null
null
[ -0.024874188005924225, -0.011809284798800945, -0.01316610723733902, 0.03565841168165207, 0.04068000987172127, 0.012126634828746319, 0.017534401267766953, 0.01563434489071369, -0.03549288585782051, -0.028197361156344414, -0.05152658745646477, 0.019199082627892494, -0.08840982615947723, -0.0...
Pretraining task diversity and the emergence of non-Bayesian in-context learning for regression
https://openreview.net/forum?id=BtAz4a5xDg
[ "Allan Raventos", "Mansheej Paul", "Feng Chen", "Surya Ganguli" ]
Poster
null
Pretrained transformers exhibit the remarkable ability of in-context learning (ICL): they can learn tasks from just a few examples provided in the prompt without updating any weights. This raises a foundational question: can ICL solve fundamentally _new_ tasks that are very different from those seen during pretraining?...
[ "in-context learning", "Bayesian inference", "transformers", "task diversity", "emergence" ]
We empirically demonstrate a task diversity threshold for the emergence of in-context learning in pretrained transformers beyond which the model can learn fundamentally new tasks in-context.
14,855
2306.15063
title_snapshot
[ -0.003432377241551876, -0.019741229712963104, -0.013859630562365055, 0.028044961392879486, 0.032750267535448074, 0.030565213412046432, 0.03096211701631546, 0.0026856001932173967, -0.025691553950309753, -0.017342349514365196, -0.049384281039237976, 0.054723478853702545, -0.05927000567317009, ...
Experiment Planning with Function Approximation
https://openreview.net/forum?id=axmY49ahVI
[ "Aldo Pacchiano", "Jonathan Lee", "Emma Brunskill" ]
Poster
null
We study the problem of experiment planning with function approximation in contextual bandit problems. In settings where there is a significant overhead to deploying adaptive algorithms---for example, when the execution of the data collection policies is required to be distributed, or a human in the loop is needed to i...
[ "regret", "model selection", "planning", "static", "lower bound" ]
null
14,853
2401.05193
title_snapshot
[ -0.024196796119213104, -0.009259463287889957, -0.034290559589862823, 0.05737508088350296, 0.04529304429888725, 0.03606536239385605, 0.03158002346754074, -0.023283081129193306, -0.03336965665221214, -0.02778327837586403, -0.020419305190443993, 0.023166369646787643, -0.04640326648950577, -0....
MIMEx: Intrinsic Rewards from Masked Input Modeling
https://openreview.net/forum?id=g1dMYenhe4
[ "Toru Lin", "Allan Jabri" ]
Poster
null
Exploring in environments with high-dimensional observations is hard. One promising approach for exploration is to use intrinsic rewards, which often boils down to estimating "novelty" of states, transitions, or trajectories with deep networks. Prior works have shown that conditional prediction objectives such as maske...
[ "reinforcement learning", "exploration", "intrinsic reward", "intrinsic motivation", "masked autoencoder" ]
null
14,852
2305.08932
title_snapshot
[ -0.011411032639443874, -0.0047934758476912975, -0.014455290511250496, 0.05161520466208458, 0.044223856180906296, 0.046670183539390564, 0.024996638298034668, 0.002705960301682353, -0.03483568876981735, -0.04932098463177681, -0.02695143222808838, 0.006456286180764437, -0.04917012155056, -0.0...
Prioritizing Samples in Reinforcement Learning with Reducible Loss
https://openreview.net/forum?id=g78QqvhnDU
[ "Shiva Kanth Sujit", "Somjit Nath", "Pedro Braga", "Samira Ebrahimi Kahou" ]
Poster
null
Most reinforcement learning algorithms take advantage of an experience replay buffer to repeatedly train on samples the agent has observed in the past. Not all samples carry the same amount of significance and simply assigning equal importance to each of the samples is a naïve strategy. In this paper, we propose a meth...
[ "reinforcement learning", "sample efficiency", "experience replay" ]
We propose a prioritization scheme based on the loss reduction potential instead of just the magnitude of the loss
14,846
2208.10483
title_snapshot
[ -0.046476900577545166, -0.03334181010723114, -0.014075692743062973, 0.06638925522565842, 0.043620746582746506, 0.01280368771404028, 0.007129826117306948, -0.002980998018756509, -0.03750673681497574, -0.04529057443141937, -0.014845607802271843, 0.02220052480697632, -0.06312983483076096, -0....
Spatially Resolved Gene Expression Prediction from Histology Images via Bi-modal Contrastive Learning
https://openreview.net/forum?id=eT1tMdAUoc
[ "Ronald Xie", "Kuan Pang", "Sai W Chung", "Catia Perciani", "Sonya MacParland", "BO WANG", "Gary Bader" ]
Poster
null
Histology imaging is an important tool in medical diagnosis and research, enabling the examination of tissue structure and composition at the microscopic level. Understanding the underlying molecular mechanisms of tissue architecture is critical in uncovering disease mechanisms and developing effective treatments.Gene ...
[ "BLEEP", "Histology", "H&E", "Gene Expression Prediction", "Spatial Transcriptomics", "Contrastive Learning" ]
null
14,824
2306.01859
title_judge
[ 0.024592671543359756, 0.010360795073211193, 0.001271645538508892, 0.009945996105670929, 0.04946395754814148, 0.040929071605205536, 0.02120143733918667, 0.0033009531907737255, 0.00033706307294778526, -0.020392145961523056, 0.016266105696558952, -0.00047079630894586444, -0.037173330783843994, ...
Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcuts
https://openreview.net/forum?id=tLTtqySDFb
[ "Emanuele Marconato", "Stefano Teso", "Antonio Vergari", "Andrea Passerini" ]
Poster
null
Neuro-Symbolic (NeSy) predictive models hold the promise of improved compliance with given constraints, systematic generalization, and interpretability, as they allow to infer labels that are consistent with some prior knowledge by reasoning over high-level concepts extracted from sub-symbolic inputs. It was recently s...
[ "Neuro-Symbolic Integration", "Trustworthy AI", "Concept Learning", "Learning Shortcuts", "Mitigation Strategies" ]
A thorough analysis of learning shortcuts acquired by Neuro-Symbolic prediction models and of potential mitigation strategies
14,823
2305.19951
title_snapshot
[ -0.03523588553071022, -0.011192682199180126, -0.02406686544418335, 0.03209870681166649, 0.06131274253129959, 0.027194973081350327, 0.044999364763498306, -0.010384203866124153, -0.04180319607257843, -0.015766246244311333, -0.037406861782073975, 0.06169208884239197, -0.0497935526072979, -0.0...
Group Robust Classification Without Any Group Information
https://openreview.net/forum?id=2OcNWFHFpk
[ "Christos Tsirigotis", "Joao Monteiro", "Pau Rodriguez", "David Vazquez", "Aaron Courville" ]
Poster
null
Empirical risk minimization (ERM) is sensitive to spurious correlations present in training data, which poses a significant risk when deploying systems trained under this paradigm in high-stake applications. While the existing literature focuses on maximizing group-balanced or worst-group accuracy, estimating these qua...
[ "out-of-distribution generalization", "robustness", "fairness", "spurious correlations", "systematic generalization", "model selection" ]
We use pretrained models to define an entirely bias-unsupervised training and validation methodology for group robust classification
14,819
2310.18555
title_snapshot
[ -0.020273733884096146, -0.0037152201402932405, 0.01720239780843258, 0.04163292422890663, -0.003206652821972966, 0.013206363655626774, 0.029165644198656082, -0.00797316525131464, -0.016621196642518044, -0.05649775639176369, -0.019422587007284164, 0.002651774790138006, -0.0828540027141571, 0...
Egocentric Planning for Scalable Embodied Task Achievement
https://openreview.net/forum?id=v0lkbp66Uw
[ "Xiaotian Liu", "Hector Palacios", "Christian Muise" ]
Poster
null
Embodied agents face significant challenges when tasked with performing actions in diverse environments, particularly in generalizing across object types and executing suitable actions to accomplish tasks. Furthermore, agents should exhibit robustness, minimizing the execution of illegal actions. In this work, we prese...
[ "Embodied AI", "High-Level Actions", "Symbolic Reasoning", "Replanning", "ALFRED Challenge", "Flexible Task Achievement", "User-Goal Understanding", "Object Types and Actions", "Perception Grounding" ]
In high-level environments, flexible embodied AI agents leverage known object types and actions for grounding perception and understanding user-goals. Our fast, iterative symbolic replanning won the ALFRED challenge
14,817
2306.01295
title_snapshot
[ -0.019569383934140205, -0.021677503362298012, -0.0009397457470186055, -0.0021000795532017946, 0.007170466240495443, 0.035669051110744476, 0.03562048450112343, -0.01729404553771019, -0.04069053381681442, -0.0268411748111248, -0.030002128332853317, 0.009758979082107544, -0.04801555722951889, ...
Lie Point Symmetry and Physics-Informed Networks
https://openreview.net/forum?id=ba4boN3W1n
[ "Tara Akhound-Sadegh", "Laurence Perreault-Levasseur", "Johannes Brandstetter", "Max Welling", "Siamak Ravanbakhsh" ]
Poster
null
Symmetries have been leveraged to improve the generalization of neural networks through different mechanisms from data augmentation to equivariant architectures. However, despite their potential, their integration into neural solvers for partial differential equations (PDEs) remains largely unexplored. We explore the i...
[ "PDE", "Lie point symmetry", "Symmetry", "Neural PDE solver", "PINNs" ]
We use Lie Point Symmetries of PDEs to Improve Physics Informed Networks.
14,815
2311.04293
title_snapshot
[ -0.05164295807480812, -0.009746579453349113, 0.02129157818853855, 0.027642739936709404, -0.000688995816744864, 0.00999247096478939, 0.009652646258473396, -0.015281720086932182, -0.0350412055850029, -0.03472860902547836, -0.0006188718252815306, -0.0393306128680706, -0.04558328911662102, 0.0...
PAC-Bayes Generalization Certificates for Learned Inductive Conformal Prediction
https://openreview.net/forum?id=URrUpcp6Qh
[ "Apoorva Sharma", "Sushant Veer", "Asher Hancock", "Heng Yang", "Marco Pavone", "Anirudha Majumdar" ]
Poster
null
Inductive Conformal Prediction (ICP) provides a practical and effective approach for equipping deep learning models with uncertainty estimates in the form of set-valued predictions which are guaranteed to contain the ground truth with high probability. Despite the appeal of this coverage guarantee, these sets may not b...
[ "Conformal Prediction", "PAC Bayes", "Generalization Theory" ]
We derive PAC-Bayes generalization bounds for learned inductive conformal prediction and show how these bounds can be used to yield efficient and valid prediction sets.
14,811
2312.04658
title_snapshot
[ -0.006154459435492754, -0.013664113357663155, -0.0019924091175198555, 0.05162334442138672, 0.03808116167783737, 0.01834544725716114, 0.017458375543355942, -0.01619310863316059, -0.015965856611728668, -0.03592384234070778, 0.0004183600249234587, 0.003992604557424784, -0.0715312510728836, 0....
Derandomized novelty detection with FDR control via conformal e-values
https://openreview.net/forum?id=toYvRJ7Zmy
[ "Meshi Bashari", "Amir Epstein", "Yaniv Romano", "Matteo Sesia" ]
Poster
null
Conformal inference provides a general distribution-free method to rigorously calibrate the output of any machine learning algorithm for novelty detection. While this approach has many strengths, it has the limitation of being randomized, in the sense that it may lead to different results when analyzing twice the same ...
[ "Conformal inference", "Derandomization", "E-values", "False discovery rate", "Out-of-distribution testing", "Testing for outliers", "Uncertainty" ]
Powerful derandomization framework for conformalized novelty detection under rigorous false discovery rate control.
14,808
2302.07294
title_snapshot
[ 0.009041841141879559, -0.020398493856191635, 0.0027464188169687986, 0.05418745428323746, 0.0652037262916565, -0.02443552017211914, 0.017772551625967026, -0.026628991588950157, -0.013820561580359936, -0.06531748175621033, -0.007015179377049208, -0.003668722230941057, -0.071757972240448, 0.0...
Adversarial Learning for Feature Shift Detection and Correction
https://openreview.net/forum?id=lBhRTO2uWf
[ "Míriam Barrabés", "Daniel Mas Montserrat", "Margarita Geleta", "Xavier Giró-i-Nieto", "Alexander G Ioannidis" ]
Poster
null
Data shift is a phenomenon present in many real-world applications, and while there are multiple methods attempting to detect shifts, the task of localizing and correcting the features originating such shifts has not been studied in depth. Feature shifts can occur in many datasets, including in multi-sensor data, where...
[ "feature shift detection", "distribution shift", "shift", "data-centric AI" ]
We introduce a framework inspired in adversarial learning to detect and correct features originating a distribution shift between datasets.
14,805
2312.04546
title_snapshot
[ 0.0062171039171516895, -0.0389828234910965, -0.028776099905371666, 0.06464049965143204, 0.06137187033891678, 0.028486549854278564, 0.029609298333525658, -0.024674993008375168, -0.017279809340834618, -0.0672801285982132, -0.03245095908641815, 0.013552660122513771, -0.07028724998235703, -0.0...
Grammar Prompting for Domain-Specific Language Generation with Large Language Models
https://openreview.net/forum?id=B4tkwuzeiY
[ "Bailin Wang", "Zi Wang", "Xuezhi Wang", "Yuan Cao", "Rif A. Saurous", "Yoon Kim" ]
Poster
null
Large language models (LLMs) can learn to perform a wide range of natural language tasks from just a handful of in-context examples. However, for generating strings from highly structured languages (e.g., semantic parsing to complex domain-specific languages), it is challenging for the LLM to generalize from just a ...
[ "semantic parsing", "large language models", "PDDL", "AI planning", "molecule generation", "data efficiency", "grammar-based learning" ]
We propose grammar prompting, a simple approach for improving the few-shot DSL generation with large language models.
14,797
2305.19234
title_snapshot
[ -0.028538204729557037, -0.002707122825086117, -0.02198476903140545, 0.00928858295083046, 0.026799453422427177, 0.017469653859734535, 0.033543795347213745, 0.007214376237243414, -0.026151396334171295, 0.0006081322790123522, -0.03240232542157173, 0.034988708794116974, -0.06056592985987663, 0...
Rethinking Gauss-Newton for learning over-parameterized models
https://openreview.net/forum?id=8Oukmqfek2
[ "Michael Arbel", "Romain Menegaux", "Pierre Wolinski" ]
Poster
null
This work studies the global convergence and implicit bias of Gauss Newton's (GN) when optimizing over-parameterized one-hidden layer networks in the mean-field regime. We first establish a global convergence result for GN in the continuous-time limit exhibiting a faster convergence rate compared to GD due to improved...
[ "implicit bias", "gauss newton" ]
global convergence and generalization of gauss-newton methods for over-parameterized one-hidden layer networks
14,789
2302.02904
title_snapshot
[ -0.04874303936958313, -0.03175933286547661, 0.030398735776543617, 0.03264329954981804, 0.020551418885588646, 0.03248393163084984, 0.02639501914381981, 0.03425370156764984, -0.03749038651585579, -0.036034367978572845, 0.0006242456147447228, 0.005499419756233692, -0.07437954843044281, -0.004...
Recovering Simultaneously Structured Data via Non-Convex Iteratively Reweighted Least Squares
https://openreview.net/forum?id=50hs53Zb3w
[ "Christian Kümmerle", "Johannes Maly" ]
Poster
null
We propose a new algorithm for the problem of recovering data that adheres to multiple, heterogenous low-dimensional structures from linear observations. Focussing on data matrices that are simultaneously row-sparse and low-rank, we propose and analyze an iteratively reweighted least squares (IRLS) algorithm that is ab...
[ "low-rank models", "sparsity", "iteratively reweighted least squares", "non-convex optimization", "quadratic convergence", "simultaneously structured data" ]
We propose a novel iteratively reweighted least squares method for the recovery of simultaneously structured data from measurements, and show its local quadratic convergence in a minimal sample complexity regime.
14,765
2306.04961
title_snapshot
[ -0.02922130934894085, -0.019334230571985245, 0.023055147379636765, 0.027272215113043785, 0.03881612792611122, 0.03409966453909874, 0.015366851352155209, -0.007782716304063797, -0.03269428387284279, -0.0465109758079052, -0.007399085909128189, -0.003218061290681362, -0.05539235100150108, -0....
SOL: Sampling-based Optimal Linear bounding of arbitrary scalar functions
https://openreview.net/forum?id=gAQCx61chN
[ "Yuriy Biktairov", "Jyotirmoy Deshmukh" ]
Poster
null
Finding tight linear bounds for activation functions in neural networks is an essential part of several state of the art neural network robustness certification tools. An activation function is an arbitrary, nonlinear, scalar function $f: \mathbb{R}^d \rightarrow \mathbb{R}$. In the existing work on robustness certif...
[ "Neural network verification", "Robustness", "Linear bounding" ]
null
14,761
null
null
[ -0.03769242763519287, 0.012955578975379467, 0.00779872527346015, 0.03127061948180199, 0.02900855429470539, 0.040591806173324585, 0.021120693534612656, -0.035228531807661057, -0.034524910151958466, -0.01823911815881729, 0.025323104113340378, 0.0006185580277815461, -0.07227974385023117, -0.0...
Compositional Sculpting of Iterative Generative Processes
https://openreview.net/forum?id=w79RtqIyoM
[ "Timur Garipov", "Sebastiaan De Peuter", "Ge Yang", "Vikas Garg", "Samuel Kaski", "Tommi S. Jaakkola" ]
Poster
null
High training costs of generative models and the need to fine-tune them for specific tasks have created a strong interest in model reuse and composition. A key challenge in composing iterative generative processes, such as GFlowNets and diffusion models, is that to realize the desired target distribution, all steps of ...
[ "generative model composition", "GFlowNets", "diffusion models", "classifier guidance", "probabilistic methods" ]
We propose a framework to compose iterative generative processes: GFlowNets and diffusion models.
14,758
2309.16115
title_snapshot
[ 0.02504500187933445, -0.020665982738137245, -0.016453783959150314, 0.044236574321985245, 0.030587857589125633, 0.04570426046848297, -0.016010036692023277, 0.034998569637537, -0.0122177479788661, -0.057576458901166916, -0.014809731394052505, -0.0407414510846138, -0.08427649736404419, 0.0159...
High-Fidelity Audio Compression with Improved RVQGAN
https://openreview.net/forum?id=qjnl1QUnFA
[ "Rithesh Kumar", "Prem Seetharaman", "Alejandro Luebs", "Ishaan Kumar", "Kundan Kumar" ]
Spotlight
null
Language models have been successfully used to model natural signals, such as images, speech, and music. A key component of these models is a high quality neural compression model that can compress high-dimensional natural signals into lower dimensional discrete tokens. To that end, we introduce a high-fidelity univers...
[ "audio generation", "audio compression", "GAN", "audio", "speech" ]
A method for neural audio compression that outperforms competing approaches in terms of audio quality, at much higher compression rates.
14,752
2306.06546
title_snapshot
[ -0.00993436761200428, -0.02285422943532467, -0.00984007865190506, 0.03618486225605011, 0.02565816044807434, 0.04255129396915436, 0.020439697429537773, -0.0075346240773797035, -0.0185176320374012, -0.06483416259288788, -0.0213578213006258, 0.02276834100484848, -0.05093453824520111, 0.019183...
A State Representation for Diminishing Rewards
https://openreview.net/forum?id=7Uix1eQZ8z
[ "Ted Moskovitz", "Samo Hromadka", "Ahmed Touati", "Diana L Borsa", "Maneesh Sahani" ]
Poster
null
A common setting in multitask reinforcement learning (RL) demands that an agent rapidly adapt to various stationary reward functions randomly sampled from a fixed distribution. In such situations, the successor representation (SR) is a popular framework which supports rapid policy evaluation by decoupling a policy's ex...
[ "reinforcement learning", "successor features", "successor representation", "neuroscience" ]
We study the problem of diminishing marginal utility in reinforcement learning, deriving a novel state representation that is required for solving such problems and applying it in both tabular and continuous domains.
14,740
2309.03710
title_snapshot
[ -0.039717283099889755, -0.01882374845445156, -0.005501355044543743, 0.031185997650027275, 0.05209112912416458, 0.03487410023808479, 0.005081622861325741, -0.014369421638548374, -0.04813176020979881, -0.048615407198667526, -0.012935778126120567, 0.01973947510123253, -0.08031205087900162, -0...
Discriminative Calibration: Check Bayesian Computation from Simulations and Flexible Classifier
https://openreview.net/forum?id=2Cmdh5z6ph
[ "Yuling Yao", "Justin Domke" ]
Poster
null
To check the accuracy of Bayesian computations, it is common to use rank-based simulation-based calibration (SBC). However, SBC has drawbacks: The test statistic is somewhat ad-hoc, interactions are difficult to examine, multiple testing is a challenge, and the resulting p-value is not a divergence metric. We propose t...
[ "simulation based calibration", "simulation based inference", "Bayesian computation", "diagnostics", "classifier two-sample test", "likelihood-free" ]
We use a classifier+statistical feature approach to diagnose Bayesian computations, applicable to MCMC, variational and simulation based inference. We provide a high-power test and a divergence estimate.
14,735
2305.14593
title_snapshot
[ 0.0059928977862000465, 0.0017697125440463424, -0.028655173256993294, 0.02659684605896473, 0.052860599011182785, 0.029347505420446396, 0.037885118275880814, -0.03232673555612564, -0.03974222391843796, -0.04379682615399361, 0.03598672151565552, 0.016925543546676636, -0.054451748728752136, -0...