arxivId
stringlengths
11
18
arxivBaseId
stringlengths
9
16
version
int64
1
19
title
stringlengths
8
207
abstract
stringlengths
87
2.34k
abstractWordCount
int64
12
315
authors
listlengths
1
3.44k
authorCount
int64
1
3.44k
firstAuthor
stringlengths
3
55
authorAffiliations
listlengths
0
184
primaryCategory
stringclasses
137 values
categories
listlengths
1
7
isCrossListed
bool
2 classes
publishedAt
timestamp[s]date
1995-10-01 00:00:00
2026-07-30 17:59:56
publishedDate
timestamp[s]date
1995-10-01 00:00:00
2026-07-30 00:00:00
updatedAt
timestamp[s]date
1995-10-01 00:00:00
2026-07-30 17:59:56
doi
stringclasses
655 values
doiUrl
stringclasses
655 values
journalRef
stringclasses
541 values
comment
stringlengths
1
463
absUrl
stringlengths
33
40
pdfUrl
stringlengths
33
40
sourceQuery
stringclasses
51 values
scrapedAt
timestamp[s]date
2026-08-02 22:16:30
2026-08-02 22:40:37
_run_id
stringclasses
90 values
_dataset_id
stringclasses
90 values
2506.13672v1
2506.13672
1
The Courage to Stop: Overcoming Sunk Cost Fallacy in Deep Reinforcement Learning
Off-policy deep reinforcement learning (RL) typically leverages replay buffers for reusing past experiences during learning. This can help improve sample efficiency when the collected data is informative and aligned with the learning objectives; when that is not the case, it can have the effect of "polluting" the repla...
158
[ "Jiashun Liu", "Johan Obando-Ceron", "Pablo Samuel Castro", "Aaron Courville", "Ling Pan" ]
5
Jiashun Liu
[]
cs.LG
[ "cs.LG" ]
false
2025-06-16T16:30:00
2025-06-16T00:00:00
2025-06-16T16:30:00
null
null
null
Proceedings of the 42nd International Conference on Machine Learning (ICML 2025)
https://arxiv.org/abs/2506.13672v1
https://arxiv.org/pdf/2506.13672v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2506.05419v1
2506.05419
1
Dream to Generalize: Zero-Shot Model-Based Reinforcement Learning for Unseen Visual Distractions
Model-based reinforcement learning (MBRL) has been used to efficiently solve vision-based control tasks in highdimensional image observations. Although recent MBRL algorithms perform well in trained observations, they fail when faced with visual distractions in observations. These task-irrelevant distractions (e.g., cl...
177
[ "Jeongsoo Ha", "Kyungsoo Kim", "Yusung Kim" ]
3
Jeongsoo Ha
[]
cs.CV
[ "cs.CV", "cs.AI" ]
true
2025-06-05T00:39:03
2025-06-05T00:00:00
2025-06-05T00:39:03
null
null
null
AAAI 2023
https://arxiv.org/abs/2506.05419v1
https://arxiv.org/pdf/2506.05419v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2506.03758v1
2506.03758
1
Scaling CrossQ with Weight Normalization
Reinforcement learning has achieved significant milestones, but sample efficiency remains a bottleneck for real-world applications. Recently, CrossQ has demonstrated state-of-the-art sample efficiency with a low update-to-data (UTD) ratio of 1. In this work, we explore CrossQ's scaling behavior with higher UTD ratios. ...
153
[ "Daniel Palenicek", "Florian Vogt", "Jan Peters" ]
3
Daniel Palenicek
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2025-06-04T09:24:17
2025-06-04T00:00:00
2025-06-04T09:24:17
null
null
null
arXiv admin note: substantial text overlap with arXiv:2502.07523
https://arxiv.org/abs/2506.03758v1
https://arxiv.org/pdf/2506.03758v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2506.01896v1
2506.01896
1
Sums and differences of sets: a further improvement over AlphaEvolve
We present a new advancement in the sum and difference of sets problem, which improves upon recent results by both DeepMind's AlphaEvolve ($θ= 1.1584$) and subsequent explicit constructions ($θ= 1.173050$). In this work, we construct a sequence of $U$ sets which in the limit establishes a new lower bound of $θ= 1.17307...
52
[ "Fan Zheng" ]
1
Fan Zheng
[]
math.CO
[ "math.CO", "math.NT" ]
true
2025-06-02T17:20:25
2025-06-02T00:00:00
2025-06-02T17:20:25
null
null
null
11 pages
https://arxiv.org/abs/2506.01896v1
https://arxiv.org/pdf/2506.01896v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2506.00920v2
2506.00920
2
Position as Probability: Self-Supervised Transformers that Think Past Their Training for Length Extrapolation
Deep sequence models typically degrade in accuracy when test sequences significantly exceed their training lengths, yet many critical tasks--such as algorithmic reasoning, multi-step arithmetic, and compositional generalization--require robust length extrapolation. We introduce PRISM, a Probabilistic Relative-position ...
152
[ "Philip Heejun Lee" ]
1
Philip Heejun Lee
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CL", "cs.NE" ]
true
2025-06-01T09:20:44
2025-06-01T00:00:00
2025-12-22T19:46:45
null
null
null
Note: v2: working paper; code, additional baselines, ablations, will follow in v3
https://arxiv.org/abs/2506.00920v2
https://arxiv.org/pdf/2506.00920v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2506.00592v1
2506.00592
1
Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn
Plasticity, or the ability of an agent to adapt to new tasks, environments, or distributions, is crucial for continual learning. In this paper, we study the loss of plasticity in deep continual RL from the lens of churn: network output variability for out-of-batch data induced by mini-batch training. We demonstrate tha...
129
[ "Hongyao Tang", "Johan Obando-Ceron", "Pablo Samuel Castro", "Aaron Courville", "Glen Berseth" ]
5
Hongyao Tang
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2025-05-31T14:58:22
2025-05-31T00:00:00
2025-05-31T14:58:22
null
null
null
Accepted to ICML 2025
https://arxiv.org/abs/2506.00592v1
https://arxiv.org/pdf/2506.00592v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.24061v2
2505.24061
2
Measure gradients, not activations! Enhancing neuronal activity in deep reinforcement learning
Deep reinforcement learning (RL) agents frequently suffer from neuronal activity loss, which impairs their ability to adapt to new data and learn continually. A common method to quantify and address this issue is the tau-dormant neuron ratio, which uses activation statistics to measure the expressive ability of neurons...
172
[ "Jiashun Liu", "Zihao Wu", "Johan Obando-Ceron", "Pablo Samuel Castro", "Aaron Courville", "Ling Pan" ]
6
Jiashun Liu
[]
cs.LG
[ "cs.LG" ]
false
2025-05-29T23:07:58
2025-05-29T00:00:00
2026-01-31T04:43:22
null
null
null
null
https://arxiv.org/abs/2505.24061v2
https://arxiv.org/pdf/2505.24061v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.21410v2
2505.21410
2
MRS: Multi-Resolution Skills for HRL Agents
Hierarchical reinforcement learning (HRL) decomposes the policy into a manager and a worker, enabling long-horizon planning but introducing a performance gap on tasks requiring agility. We identify a root cause: in subgoal-based HRL, the manager's goal representation is typically learned without constraints on reachabi...
149
[ "Shashank Sharma", "Janina Hoffmann", "Vinay Namboodiri" ]
3
Shashank Sharma
[]
cs.AI
[ "cs.AI", "cs.LG", "cs.RO" ]
true
2025-05-27T16:38:55
2025-05-27T00:00:00
2026-04-21T04:20:38
null
null
null
null
https://arxiv.org/abs/2505.21410v2
https://arxiv.org/pdf/2505.21410v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.18899v1
2505.18899
1
Beyond Domain Randomization: Event-Inspired Perception for Visually Robust Adversarial Imitation from Videos
Imitation from videos often fails when expert demonstrations and learner environments exhibit domain shifts, such as discrepancies in lighting, color, or texture. While visual randomization partially addresses this problem by augmenting training data, it remains computationally intensive and inherently reactive, strugg...
186
[ "Andrea Ramazzina", "Vittorio Giammarino", "Matteo El-Hariry", "Mario Bijelic" ]
4
Andrea Ramazzina
[]
cs.CV
[ "cs.CV", "cs.LG", "cs.RO" ]
true
2025-05-24T23:12:23
2025-05-24T00:00:00
2025-05-24T23:12:23
null
null
null
null
https://arxiv.org/abs/2505.18899v1
https://arxiv.org/pdf/2505.18899v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.16787v3
2505.16787
3
Enter the Void - Planning to Seek Entropy When Reward is Scarce
Model-based reinforcement learning (MBRL) offers an intuitive way to increase the sample efficiency of model-free RL methods by simultaneously training a world model that learns to predict the future. These models constitute the large majority of training compute and time and they are subsequently used to train actors ...
245
[ "Ashish Sundar", "Chunbo Luo", "Xiaoyang Wang" ]
3
Ashish Sundar
[]
cs.AI
[ "cs.AI" ]
false
2025-05-22T15:28:50
2025-05-22T00:00:00
2025-12-18T00:27:12
null
null
null
10 pages without appendix, 15 Figures, preprint
https://arxiv.org/abs/2505.16787v3
https://arxiv.org/pdf/2505.16787v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.16105v1
2505.16105
1
Sums and differences of sets (improvement over AlphaEvolve)
On May 14, 2025, DeepMind announced that AlphaEvolve, a large language model applied to a set of mathematical problems, had matched or exceeded the best known bounds on several problems. In the case of the sum and difference of sets problem, AlphaEvolve, using a set of $54265$ integers, improved the known lower bound o...
92
[ "Robert Gerbicz" ]
1
Robert Gerbicz
[]
math.NT
[ "math.NT" ]
false
2025-05-22T01:19:35
2025-05-22T00:00:00
2025-05-22T01:19:35
null
null
null
4 pages
https://arxiv.org/abs/2505.16105v1
https://arxiv.org/pdf/2505.16105v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.14534v1
2505.14534
1
Lessons from Defending Gemini Against Indirect Prompt Injections
Gemini is increasingly used to perform tasks on behalf of users, where function-calling and tool-use capabilities enable the model to access user data. Some tools, however, require access to untrusted data introducing risk. Adversaries can embed malicious instructions in untrusted data which cause the model to deviate ...
130
[ "Chongyang Shi", "Sharon Lin", "Shuang Song", "Jamie Hayes", "Ilia Shumailov", "Itay Yona", "Juliette Pluto", "Aneesh Pappu", "Christopher A. Choquette-Choo", "Milad Nasr", "Chawin Sitawarin", "Gena Gibson", "Andreas Terzis", "John \"Four\" Flynn" ]
14
Chongyang Shi
[]
cs.CR
[ "cs.CR", "cs.LG" ]
true
2025-05-20T15:54:45
2025-05-20T00:00:00
2025-05-20T15:54:45
null
null
null
null
https://arxiv.org/abs/2505.14534v1
https://arxiv.org/pdf/2505.14534v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.10947v4
2505.10947
4
Certifying Stability of Reinforcement Learning Policies using Generalized Lyapunov Functions
Establishing stability certificates for closed-loop systems under reinforcement learning (RL) policies is essential to move beyond empirical performance and offer guarantees of system behavior. Classical Lyapunov methods require a strict stepwise decrease in the Lyapunov function but such certificates are difficult to ...
246
[ "Kehan Long", "Jorge Cortés", "Nikolay Atanasov" ]
3
Kehan Long
[]
cs.LG
[ "cs.LG", "cs.RO", "eess.SY", "math.OC" ]
true
2025-05-16T07:36:40
2025-05-16T00:00:00
2026-01-12T01:08:26
null
null
null
NeurIPS 2025
https://arxiv.org/abs/2505.10947v4
https://arxiv.org/pdf/2505.10947v4
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.09959v1
2505.09959
1
Approximated Behavioral Metric-based State Projection for Federated Reinforcement Learning
Federated reinforcement learning (FRL) methods usually share the encrypted local state or policy information and help each client to learn from others while preserving everyone's privacy. In this work, we propose that sharing the approximated behavior metric-based state projection function is a promising way to enhance...
116
[ "Zengxia Guo", "Bohui An", "Zhongqi Lu" ]
3
Zengxia Guo
[]
cs.LG
[ "cs.LG" ]
false
2025-05-15T04:41:21
2025-05-15T00:00:00
2025-05-15T04:41:21
null
null
null
null
https://arxiv.org/abs/2505.09959v1
https://arxiv.org/pdf/2505.09959v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.07956v1
2505.07956
1
Symbolic Regression with Multimodal Large Language Models and Kolmogorov Arnold Networks
We present a novel approach to symbolic regression using vision-capable large language models (LLMs) and the ideas behind Google DeepMind's Funsearch. The LLM is given a plot of a univariate function and tasked with proposing an ansatz for that function. The free parameters of the ansatz are fitted using standard numer...
149
[ "Thomas R. Harvey", "Fabian Ruehle", "Kit Fraser-Taliente", "James Halverson" ]
4
Thomas R. Harvey
[]
cs.LG
[ "cs.LG", "cs.NE", "cs.SC" ]
true
2025-05-12T18:00:41
2025-05-12T00:00:00
2025-05-12T18:00:41
null
null
null
null
https://arxiv.org/abs/2505.07956v1
https://arxiv.org/pdf/2505.07956v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.00681v1
2505.00681
1
MINERVA: Evaluating Complex Video Reasoning
Multimodal LLMs are turning their focus to video benchmarks, however most video benchmarks only provide outcome supervision, with no intermediate or interpretable reasoning steps. This makes it challenging to assess if models are truly able to combine perceptual and temporal information to reason about videos, or simpl...
198
[ "Arsha Nagrani", "Sachit Menon", "Ahmet Iscen", "Shyamal Buch", "Ramin Mehran", "Nilpa Jha", "Anja Hauth", "Yukun Zhu", "Carl Vondrick", "Mikhail Sirotenko", "Cordelia Schmid", "Tobias Weyand" ]
12
Arsha Nagrani
[]
cs.LG
[ "cs.LG", "cs.CV" ]
true
2025-05-01T17:41:49
2025-05-01T00:00:00
2025-05-01T17:41:49
null
null
null
null
https://arxiv.org/abs/2505.00681v1
https://arxiv.org/pdf/2505.00681v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.00663v1
2505.00663
1
Wasserstein Policy Optimization
We introduce Wasserstein Policy Optimization (WPO), an actor-critic algorithm for reinforcement learning in continuous action spaces. WPO can be derived as an approximation to Wasserstein gradient flow over the space of all policies projected into a finite-dimensional parameter space (e.g., the weights of a neural netw...
131
[ "David Pfau", "Ian Davies", "Diana Borsa", "Joao G. M. Araujo", "Brendan Tracey", "Hado van Hasselt" ]
6
David Pfau
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2025-05-01T17:07:01
2025-05-01T00:00:00
2025-05-01T17:07:01
null
null
null
Accepted to ICML 2025
https://arxiv.org/abs/2505.00663v1
https://arxiv.org/pdf/2505.00663v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2505.00174v2
2505.00174
2
Real-World Gaps in AI Governance Research
Drawing on 1,178 safety and reliability papers from 9,439 generative AI papers (January 2020 - March 2025), we compare research outputs of leading AI companies (Anthropic, Google DeepMind, Meta, Microsoft, and OpenAI) and AI universities (CMU, MIT, NYU, Stanford, UC Berkeley, and University of Washington). We find that...
123
[ "Ilan Strauss", "Isobel Moure", "Tim O'Reilly", "Sruly Rosenblat" ]
4
Ilan Strauss
[]
cs.AI
[ "cs.AI" ]
false
2025-04-30T20:44:42
2025-04-30T00:00:00
2025-05-05T21:12:46
10.35650/AIDP.4112.d.2025
https://doi.org/10.35650/AIDP.4112.d.2025
null
Corrected a previous error: replaced 'underrepresented in Academic AI research' with the intended phrase 'underrepresented in Corporate AI research'
https://arxiv.org/abs/2505.00174v2
https://arxiv.org/pdf/2505.00174v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2504.15181v2
2504.15181
2
Mapping Industry Practices to the EU AI Act's GPAI Code of Practice Safety and Security Measures
This report provides a detailed comparison between the Safety and Security measures proposed in the EU AI Act's General-Purpose AI (GPAI) Code of Practice (Third Draft) and the current commitments and practices voluntarily adopted by leading AI companies. As the EU moves toward enforcing binding obligations for GPAI mo...
194
[ "Lily Stelling", "Mick Yang", "Rokas Gipiškis", "Leon Staufer", "Ze Shen Chin", "Siméon Campos", "Ariel Gil", "Michael Chen" ]
8
Lily Stelling
[]
cs.CY
[ "cs.CY", "cs.AI" ]
true
2025-04-21T15:44:01
2025-04-21T00:00:00
2025-07-22T19:27:15
null
null
null
166 pages, the Oxford Martin AI Governance Initiative
https://arxiv.org/abs/2504.15181v2
https://arxiv.org/pdf/2504.15181v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2504.11844v1
2504.11844
1
Evaluating the Goal-Directedness of Large Language Models
To what extent do LLMs use their capabilities towards their given goal? We take this as a measure of their goal-directedness. We evaluate goal-directedness on tasks that require information gathering, cognitive effort, and plan execution, where we use subtasks to infer each model's relevant capabilities. Our evaluation...
105
[ "Tom Everitt", "Cristina Garbacea", "Alexis Bellot", "Jonathan Richens", "Henry Papadatos", "Siméon Campos", "Rohin Shah" ]
7
Tom Everitt
[]
cs.AI
[ "cs.AI", "cs.CL", "cs.LG" ]
true
2025-04-16T08:07:08
2025-04-16T00:00:00
2025-04-16T08:07:08
null
null
null
null
https://arxiv.org/abs/2504.11844v1
https://arxiv.org/pdf/2504.11844v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2504.05074v1
2504.05074
1
On the Performance of an Explainable Language Model on PubMedQA
Large language models (LLMs) have shown significant abilities in retrieving medical knowledge, reasoning over it and answering medical questions comparably to physicians. However, these models are not interpretable, hallucinate, are difficult to maintain and require enormous compute resources for training and inference...
145
[ "Venkat Srinivasan", "Vishaal Jatav", "Anushka Chandrababu", "Geetika Sharma" ]
4
Venkat Srinivasan
[]
cs.CL
[ "cs.CL" ]
false
2025-04-07T13:42:02
2025-04-07T00:00:00
2025-04-07T13:42:02
null
null
null
Working Paper
https://arxiv.org/abs/2504.05074v1
https://arxiv.org/pdf/2504.05074v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2504.03970v2
2504.03970
2
VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models
We introduce VideoComp, a benchmark and learning framework for advancing video-text compositionality understanding, aimed at improving vision-language models (VLMs) in fine-grained temporal alignment. Unlike existing benchmarks focused on static image-text compositionality or isolated single-event videos, our benchmark...
188
[ "Dahun Kim", "AJ Piergiovanni", "Ganesh Mallya", "Anelia Angelova" ]
4
Dahun Kim
[]
cs.CV
[ "cs.CV", "cs.AI", "cs.CL", "cs.IR" ]
true
2025-04-04T22:24:30
2025-04-04T00:00:00
2025-04-10T10:41:20
null
null
null
CVPR 2025, project page at https://github.com/google-deepmind/video_comp
https://arxiv.org/abs/2504.03970v2
https://arxiv.org/pdf/2504.03970v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.23058v1
2503.23058
1
Statistical complexity of software systems represented as multi-layer networks
Software systems are expansive, exhibiting behaviors characteristic of complex systems, such as self-organization and emergence. These systems, highlighted by advancements in Large Language Models (LLMs) and other AI applications developed by entities like DeepMind and OpenAI showcase remarkable properties. Despite the...
169
[ "Jan Žižka" ]
1
Jan Žižka
[]
cs.SE
[ "cs.SE" ]
false
2025-03-29T12:33:52
2025-03-29T00:00:00
2025-03-29T12:33:52
null
null
null
null
https://arxiv.org/abs/2503.23058v1
https://arxiv.org/pdf/2503.23058v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.22674v2
2503.22674
2
QuestBench: Can LLMs ask the right question to acquire information in reasoning tasks?
Large language models (LLMs) have shown impressive performance on reasoning benchmarks like math and logic. While many works have largely assumed well-defined tasks, real-world queries are often underspecified and only solvable by acquiring missing information. We formalize this information-gathering problem as a const...
203
[ "Belinda Z. Li", "Been Kim", "Zi Wang" ]
3
Belinda Z. Li
[]
cs.AI
[ "cs.AI", "cs.CL", "cs.LG" ]
true
2025-03-28T17:58:40
2025-03-28T00:00:00
2025-10-24T22:18:36
null
null
39th Conference on Neural Information Processing Systems (NeurIPS 2025) Track on Datasets and Benchmarks
Code and dataset are available at \url{https://github.com/google-deepmind/questbench}
https://arxiv.org/abs/2503.22674v2
https://arxiv.org/pdf/2503.22674v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.18234v2
2503.18234
2
KEA: Keeping Exploration Alive by Proactively Coordinating Exploration Strategies
Soft Actor-Critic (SAC) has achieved notable success in continuous control tasks but struggles in sparse reward settings, where infrequent rewards make efficient exploration challenging. While novelty-based exploration methods address this issue by encouraging the agent to explore novel states, they are not trivial to ...
179
[ "Shih-Min Yang", "Martin Magnusson", "Johannes A. Stork", "Todor Stoyanov" ]
4
Shih-Min Yang
[]
cs.LG
[ "cs.LG" ]
false
2025-03-23T23:00:05
2025-03-23T00:00:00
2025-06-07T17:23:47
null
null
null
Accepted to ICML 2025
https://arxiv.org/abs/2503.18234v2
https://arxiv.org/pdf/2503.18234v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.13445v3
2503.13445
3
Verbosity Tradeoffs and the Impact of Scale on the Faithfulness of LLM Self-Explanations
When asked to explain their decisions, LLMs can often give explanations which sound plausible to humans. But are these explanations faithful, i.e. do they convey the factors actually responsible for the decision? In this work, we analyse counterfactual faithfulness across 75 models from 13 families. We analyze the trad...
152
[ "Noah Y. Siegel", "Nicolas Heess", "Maria Perez-Ortiz", "Oana-Maria Camburu" ]
4
Noah Y. Siegel
[]
cs.CL
[ "cs.CL", "cs.AI" ]
true
2025-03-17T17:59:39
2025-03-17T00:00:00
2026-07-01T13:37:01
null
null
null
ICLR 2026 Workshop on Principled Design for Trustworthy AI - Interpretability, Robustness, and Safety across Modalities 67 pages, 13 figures
https://arxiv.org/abs/2503.13445v3
https://arxiv.org/pdf/2503.13445v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.17378v2
2503.17378
2
Large language model-powered AI systems achieve self-replication with no human intervention
Self-replication with no human intervention is broadly recognized as one of the principal red lines associated with frontier AI systems. While leading corporations such as OpenAI and Google DeepMind have assessed GPT-o3-mini and Gemini on replication-related tasks and concluded that these systems pose a minimal risk re...
239
[ "Xudong Pan", "Jiarun Dai", "Yihe Fan", "Minyuan Luo", "Changyi Li", "Min Yang" ]
6
Xudong Pan
[]
cs.AI
[ "cs.AI", "cs.CR", "cs.CY", "cs.ET", "cs.MA" ]
true
2025-03-14T14:44:27
2025-03-14T00:00:00
2025-03-25T13:38:18
null
null
null
Work in progress
https://arxiv.org/abs/2503.17378v2
https://arxiv.org/pdf/2503.17378v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.05934v3
2503.05934
3
Improving Merge Sort and Quick Sort Performance by Utilizing Alphadev's Sorting Networks as Base Cases
Recent work by Google DeepMind introduced assembly-optimized sorting networks that achieve faster performance for small fixed-size arrays (3-8). In this research, we investigate the integration of these networks as base cases in classical divide-and-conquer sorting algorithms, specifically Merge Sort and Quick Sort, to...
210
[ "Anas Gamal Aly", "Anders E. Jensen", "Hala ElAarag" ]
3
Anas Gamal Aly
[]
cs.DS
[ "cs.DS", "cs.CC" ]
true
2025-03-07T21:08:42
2025-03-07T00:00:00
2025-03-16T17:49:54
10.1145/3696673.3723083
https://doi.org/10.1145/3696673.3723083
Proc. ACM Southeast Conf. (ACMSE 2025) (2025) 239-244
© Anas Gamal Aly, Anders E. Jensen, and Hala ElAarag, 2025. This is the authors' version of the work. It is posted here for your personal use. Not for redistribution. The definitive Version of Record will be published in Proceedings of the 2025 ACM Southeast Conference (ACMSE 2025)
https://arxiv.org/abs/2503.05934v3
https://arxiv.org/pdf/2503.05934v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.04666v1
2503.04666
1
What Are You Doing? A Closer Look at Controllable Human Video Generation
High-quality benchmarks are crucial for driving progress in machine learning research. However, despite the growing interest in video generation, there is no comprehensive dataset to evaluate human generation. Humans can perform a wide variety of actions and interactions, but existing datasets, like TikTok and TED-Talk...
180
[ "Emanuele Bugliarello", "Anurag Arnab", "Roni Paiss", "Pieter-Jan Kindermans", "Cordelia Schmid" ]
5
Emanuele Bugliarello
[]
cs.CV
[ "cs.CV" ]
false
2025-03-06T17:59:29
2025-03-06T00:00:00
2025-03-06T17:59:29
null
null
null
null
https://arxiv.org/abs/2503.04666v1
https://arxiv.org/pdf/2503.04666v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.00713v3
2503.00713
3
Spiking World Model with Multi-Compartment Neurons for Model-based Reinforcement Learning
Brain-inspired spiking neural networks (SNNs) have garnered significant research attention in algorithm design and perception applications. However, their potential in the decision-making domain, particularly in model-based reinforcement learning, remains underexplored. The difficulty lies in the need for spiking neuro...
210
[ "Yinqian Sun", "Feifei Zhao", "Mingyang Lv", "Yi Zeng" ]
4
Yinqian Sun
[]
cs.NE
[ "cs.NE" ]
false
2025-03-02T03:40:10
2025-03-02T00:00:00
2025-03-19T09:47:14
null
null
null
null
https://arxiv.org/abs/2503.00713v3
https://arxiv.org/pdf/2503.00713v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.19187v2
2502.19187
2
BIG-Bench Extra Hard
Large language models (LLMs) are increasingly deployed in everyday applications, demanding robust general reasoning capabilities and diverse reasoning skillset. However, current LLM reasoning benchmarks predominantly focus on mathematical and coding abilities, leaving a gap in evaluating broader reasoning proficiencies...
209
[ "Mehran Kazemi", "Bahare Fatemi", "Hritik Bansal", "John Palowitch", "Chrysovalantis Anastasiou", "Sanket Vaibhav Mehta", "Lalit K. Jain", "Virginia Aglietti", "Disha Jindal", "Peter Chen", "Nishanth Dikkala", "Gladys Tyen", "Xin Liu", "Uri Shalit", "Silvia Chiappa", "Kate Olszewska", ...
20
Mehran Kazemi
[]
cs.CL
[ "cs.CL" ]
false
2025-02-26T14:50:50
2025-02-26T00:00:00
2025-05-06T15:11:32
null
null
null
null
https://arxiv.org/abs/2502.19187v2
https://arxiv.org/pdf/2502.19187v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2503.05748v1
2503.05748
1
Alignment, Agency and Autonomy in Frontier AI: A Systems Engineering Perspective
As artificial intelligence scales, the concepts of alignment, agency, and autonomy have become central to AI safety, governance, and control. However, even in human contexts, these terms lack universal definitions, varying across disciplines such as philosophy, psychology, law, computer science, mathematics, and politi...
172
[ "Krti Tallam" ]
1
Krti Tallam
[]
cs.CY
[ "cs.CY", "cs.AI" ]
true
2025-02-20T21:37:20
2025-02-20T00:00:00
2025-02-20T21:37:20
null
null
null
null
https://arxiv.org/abs/2503.05748v1
https://arxiv.org/pdf/2503.05748v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.13581v3
2502.13581
3
ActionPiece: Contextually Tokenizing Action Sequences for Generative Recommendation
Generative recommendation (GR) is an emerging paradigm where user actions are tokenized into discrete token patterns and autoregressively generated as predictions. However, existing GR models tokenize each action independently, assigning the same fixed tokens to identical actions across all sequences without considerin...
152
[ "Yupeng Hou", "Jianmo Ni", "Zhankui He", "Noveen Sachdeva", "Wang-Cheng Kang", "Ed H. Chi", "Julian McAuley", "Derek Zhiyuan Cheng" ]
8
Yupeng Hou
[]
cs.IR
[ "cs.IR", "cs.LG" ]
true
2025-02-19T09:45:29
2025-02-19T00:00:00
2025-08-15T23:44:20
null
null
null
ICML 2025 (Spotlight)
https://arxiv.org/abs/2502.13581v3
https://arxiv.org/pdf/2502.13581v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.11234v2
2502.11234
2
MaskFlow: Discrete Flows For Flexible and Efficient Long Video Generation
Generating long, high-quality videos remains a challenge due to the complex interplay of spatial and temporal dynamics and hardware limitations. In this work, we introduce MaskFlow, a unified video generation framework that combines discrete representations with flow-matching to enable efficient generation of high-qual...
159
[ "Michael Fuest", "Vincent Tao Hu", "Björn Ommer" ]
3
Michael Fuest
[]
cs.CV
[ "cs.CV" ]
false
2025-02-16T18:59:11
2025-02-16T00:00:00
2025-03-12T16:27:37
null
null
null
Project page: https://compvis.github.io/maskflow/
https://arxiv.org/abs/2502.11234v2
https://arxiv.org/pdf/2502.11234v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.10303v2
2502.10303
2
Reinforcement Learning in Strategy-Based and Atari Games: A Review of Google DeepMinds Innovations
Reinforcement Learning (RL) has been widely used in many applications, particularly in gaming, which serves as an excellent training ground for AI models. Google DeepMind has pioneered innovations in this field, employing reinforcement learning algorithms, including model-based, model-free, and deep Q-network approache...
173
[ "Abdelrhman Shaheen", "Anas Badr", "Ali Abohendy", "Hatem Alsaadawy", "Nadine Alsayad", "Ehab H. El-Shazly" ]
6
Abdelrhman Shaheen
[]
cs.AI
[ "cs.AI", "cs.GT" ]
true
2025-02-14T17:06:34
2025-02-14T00:00:00
2026-02-11T09:17:09
null
null
null
null
https://arxiv.org/abs/2502.10303v2
https://arxiv.org/pdf/2502.10303v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.07523v2
2502.07523
2
Scaling Off-Policy Reinforcement Learning with Batch and Weight Normalization
Reinforcement learning has achieved significant milestones, but sample efficiency remains a bottleneck for real-world applications. Recently, CrossQ has demonstrated state-of-the-art sample efficiency with a low update-to-data (UTD) ratio of 1. In this work, we explore CrossQ's scaling behavior with higher UTD ratios. ...
150
[ "Daniel Palenicek", "Florian Vogt", "Joe Watson", "Jan Peters" ]
4
Daniel Palenicek
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2025-02-11T12:55:32
2025-02-11T00:00:00
2025-05-22T12:39:27
null
null
null
null
https://arxiv.org/abs/2502.07523v2
https://arxiv.org/pdf/2502.07523v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.05932v2
2502.05932
2
Skill Expansion and Composition in Parameter Space
Humans excel at reusing prior knowledge to address new challenges and developing skills while solving problems. This paradigm becomes increasingly popular in the development of autonomous agents, as it develops systems that can self-evolve in response to new challenges like human beings. However, previous methods suffe...
219
[ "Tenglong Liu", "Jianxiong Li", "Yinan Zheng", "Haoyi Niu", "Yixing Lan", "Xin Xu", "Xianyuan Zhan" ]
7
Tenglong Liu
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.RO" ]
true
2025-02-09T15:22:38
2025-02-09T00:00:00
2025-03-16T11:57:19
null
null
null
ICLR 2025, 37 pages
https://arxiv.org/abs/2502.05932v2
https://arxiv.org/pdf/2502.05932v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.05555v1
2502.05555
1
Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder
While Reinforcement Learning (RL) agents can successfully learn to handle complex tasks, effectively generalizing acquired skills to unfamiliar settings remains a challenge. One of the reasons behind this is the visual encoders used are task-dependent, preventing effective feature extraction in different settings. To a...
208
[ "Yuhan Zhang", "Guoqing Ma", "Guangfu Hao", "Liangxuan Guo", "Yang Chen", "Shan Yu" ]
6
Yuhan Zhang
[]
cs.CV
[ "cs.CV" ]
false
2025-02-08T12:57:02
2025-02-08T00:00:00
2025-02-08T12:57:02
null
null
null
Accepted by AAAI 2025
https://arxiv.org/abs/2502.05555v1
https://arxiv.org/pdf/2502.05555v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.04327v2
2502.04327
2
Value-Based Deep RL Scales Predictably
Scaling data and compute is critical to the success of modern ML. However, scaling demands predictability: we want methods to not only perform well with more compute or data, but also have their performance be predictable from small-scale runs, without running the large-scale experiment. In this paper, we show that val...
197
[ "Oleh Rybkin", "Michal Nauman", "Preston Fu", "Charlie Snell", "Pieter Abbeel", "Sergey Levine", "Aviral Kumar" ]
7
Oleh Rybkin
[]
cs.LG
[ "cs.LG" ]
false
2025-02-06T18:59:47
2025-02-06T00:00:00
2025-07-25T01:10:02
null
null
null
ICML 2025
https://arxiv.org/abs/2502.04327v2
https://arxiv.org/pdf/2502.04327v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.03544v3
2502.03544
3
Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2
We present AlphaGeometry2 (AG2), a significantly improved version of AlphaGeometry introduced in (Trinh et al., 2024), which has now surpassed an average gold medalist in solving Olympiad geometry problems. To achieve this, we first extend the original AlphaGeometry language to tackle problems involving movements of ob...
196
[ "Yuri Chervonyi", "Trieu H. Trinh", "Miroslav Olšák", "Xiaomeng Yang", "Hoang Nguyen", "Marcelo Menegali", "Junehyuk Jung", "Junsu Kim", "Vikas Verma", "Quoc V. Le", "Thang Luong" ]
11
Yuri Chervonyi
[]
cs.AI
[ "cs.AI", "cs.LG" ]
true
2025-02-05T19:02:03
2025-02-05T00:00:00
2025-12-08T21:34:52
null
null
null
28 pages, 16 figures. V2: Clarified abstract, rewritten introduction, updated results on diagram generation, added acknowledgement section. V3: Added clarifications and a new section "Inequality rules", re-organized sections, added code link, now 34 pages
https://arxiv.org/abs/2502.03544v3
https://arxiv.org/pdf/2502.03544v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2502.00802v1
2502.00802
1
Fisher-Guided Selective Forgetting: Mitigating The Primacy Bias in Deep Reinforcement Learning
Deep Reinforcement Learning (DRL) systems often tend to overfit to early experiences, a phenomenon known as the primacy bias (PB). This bias can severely hinder learning efficiency and final performance, particularly in complex environments. This paper presents a comprehensive investigation of PB through the lens of th...
175
[ "Massimiliano Falzari", "Matthia Sabatelli" ]
2
Massimiliano Falzari
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2025-02-02T13:54:47
2025-02-02T00:00:00
2025-02-02T13:54:47
null
null
null
null
https://arxiv.org/abs/2502.00802v1
https://arxiv.org/pdf/2502.00802v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2501.12149v1
2501.12149
1
On the practical applicability of modern DFT functionals for chemical computations. Case study of DM21 applicability for geometry optimization
Density functional theory (DFT) is probably the most promising approach for quantum chemistry calculations considering its good balance between calculations precision and speed. In recent years, several neural network-based functionals have been developed for exchange-correlation energy approximation in DFT, DM21 devel...
144
[ "Kirill Kulaev", "Alexander Ryabov", "Michael Medvedev", "Evgeny Burnaev", "Vladimir Vanovskiy" ]
5
Kirill Kulaev
[]
physics.comp-ph
[ "physics.comp-ph", "cond-mat.mtrl-sci", "cs.AI" ]
true
2025-01-21T14:01:06
2025-01-21T00:00:00
2025-01-21T14:01:06
null
null
null
null
https://arxiv.org/abs/2501.12149v1
https://arxiv.org/pdf/2501.12149v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2501.09038v3
2501.09038
3
Do generative video models understand physical principles?
AI video generation is undergoing a revolution, with quality and realism advancing rapidly. These advances have led to a passionate scientific debate: Do video models learn "world models" that discover laws of physics -- or, alternatively, are they merely sophisticated pixel predictors that achieve visual realism witho...
170
[ "Saman Motamed", "Laura Culp", "Kevin Swersky", "Priyank Jaini", "Robert Geirhos" ]
5
Saman Motamed
[]
cs.CV
[ "cs.CV", "cs.AI", "cs.GR", "cs.LG" ]
true
2025-01-14T20:59:37
2025-01-14T00:00:00
2025-02-27T15:10:51
null
null
null
null
https://arxiv.org/abs/2501.09038v3
https://arxiv.org/pdf/2501.09038v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.18350v1
2412.18350
1
Learning Generalized Residual Exchange-Correlation-Uncertain Functional for Density Functional Theory
Density Functional Theory (DFT) stands as a widely used and efficient approach for addressing the many-electron Schrödinger equation across various domains such as physics, chemistry, and biology. However, a core challenge that persists over the long term pertains to refining the exchange-correlation (XC) approximation...
249
[ "Sizhuo Jin", "Shuo Chen", "Jianjun Qian", "Ying Tai", "Jun Li" ]
5
Sizhuo Jin
[]
cs.CE
[ "cs.CE" ]
false
2024-12-24T11:24:20
2024-12-24T00:00:00
2024-12-24T11:24:20
null
null
null
Accepted by AAAI2025
https://arxiv.org/abs/2412.18350v1
https://arxiv.org/pdf/2412.18350v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.17069v1
2412.17069
1
Optimizing Data Curation through Spectral Analysis and Joint Batch Selection (SALN)
In modern deep learning models, long training times and large datasets present significant challenges to both efficiency and scalability. Effective data curation and sample selection are crucial for optimizing the training process of deep neural networks. This paper introduces SALN, a method designed to prioritize and ...
154
[ "Mohammadreza Sharifi" ]
1
Mohammadreza Sharifi
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-12-22T15:38:36
2024-12-22T00:00:00
2024-12-22T15:38:36
null
null
null
This paper was presented at Machine Learning Knowledge Discovery (MLKD2024) conference at Amirkabir University of Technology
https://arxiv.org/abs/2412.17069v1
https://arxiv.org/pdf/2412.17069v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.15212v2
2412.15212
2
Scaling 4D Representations
Scaling has not yet been convincingly demonstrated for pure self-supervised learning from video. However, prior work has focused evaluations on semantic-related tasks $\unicode{x2013}$ action classification, ImageNet classification, etc. In this paper we focus on evaluating self-supervised learning on non-semantic visi...
134
[ "João Carreira", "Dilara Gokay", "Michael King", "Chuhan Zhang", "Ignacio Rocco", "Aravindh Mahendran", "Thomas Albert Keck", "Joseph Heyward", "Skanda Koppula", "Etienne Pot", "Goker Erdogan", "Yana Hasson", "Yi Yang", "Klaus Greff", "Guillaume Le Moing", "Sjoerd van Steenkiste", "D...
35
João Carreira
[]
cs.CV
[ "cs.CV", "cs.AI", "cs.LG" ]
true
2024-12-19T18:59:51
2024-12-19T00:00:00
2025-07-09T16:58:07
null
null
null
null
https://arxiv.org/abs/2412.15212v2
https://arxiv.org/pdf/2412.15212v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.14933v1
2412.14933
1
Cirbo: A New Tool for Boolean Circuit Analysis and Synthesis
We present an open-source tool for manipulating Boolean circuits. It implements efficient algorithms, both existing and novel, for a rich variety of frequently used circuit tasks such as satisfiability, synthesis, and minimization. We tested the tool on a wide range of practically relevant circuits (computing, in parti...
119
[ "Daniil Averkov", "Tatiana Belova", "Gregory Emdin", "Mikhail Goncharov", "Viktoriia Krivogornitsyna", "Alexander S. Kulikov", "Fedor Kurmazov", "Daniil Levtsov", "Georgie Levtsov", "Vsevolod Vaskin", "Aleksey Vorobiev" ]
11
Daniil Averkov
[]
cs.LO
[ "cs.LO", "cs.AI" ]
true
2024-12-19T15:10:31
2024-12-19T00:00:00
2024-12-19T15:10:31
null
null
null
To appear in AAAI 2025
https://arxiv.org/abs/2412.14933v1
https://arxiv.org/pdf/2412.14933v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.14312v3
2412.14312
3
Stealing That Free Lunch: Exposing the Limits of Dyna-Style Reinforcement Learning
Dyna-style off-policy model-based reinforcement learning (DMBRL) algorithms are a family of techniques for generating synthetic state transition data and thereby enhancing the sample efficiency of off-policy RL algorithms. This paper identifies and investigates a surprising performance gap observed when applying DMBRL ...
164
[ "Brett Barkley", "David Fridovich-Keil" ]
2
Brett Barkley
[]
cs.LG
[ "cs.LG" ]
false
2024-12-18T20:25:04
2024-12-18T00:00:00
2025-06-20T18:24:06
null
null
null
Accepted to ICML 2025
https://arxiv.org/abs/2412.14312v3
https://arxiv.org/pdf/2412.14312v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.14294v2
2412.14294
2
TRecViT: A Recurrent Video Transformer
We propose a novel block for \emph{causal} video modelling. It relies on a time-space-channel factorisation with dedicated blocks for each dimension: gated linear recurrent units (LRUs) perform information mixing over time, self-attention layers perform mixing over space, and MLPs over channels. The resulting architect...
157
[ "Viorica Pătrăucean", "Xu Owen He", "Joseph Heyward", "Chuhan Zhang", "Mehdi S. M. Sajjadi", "George-Cristian Muraru", "Artem Zholus", "Mahdi Karami", "Ross Goroshin", "Yutian Chen", "Simon Osindero", "João Carreira", "Razvan Pascanu" ]
13
Viorica Pătrăucean
[]
cs.CV
[ "cs.CV", "cs.LG" ]
true
2024-12-18T19:44:30
2024-12-18T00:00:00
2026-02-15T19:36:47
null
null
null
null
https://arxiv.org/abs/2412.14294v2
https://arxiv.org/pdf/2412.14294v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.12024v1
2412.12024
1
Learning to Navigate in Mazes with Novel Layouts using Abstract Top-down Maps
Learning navigation capabilities in different environments has long been one of the major challenges in decision-making. In this work, we focus on zero-shot navigation ability using given abstract $2$-D top-down maps. Like human navigation by reading a paper map, the agent reads the map as an image when navigating in a...
123
[ "Linfeng Zhao", "Lawson L. S. Wong" ]
2
Linfeng Zhao
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.RO" ]
true
2024-12-16T17:51:09
2024-12-16T00:00:00
2024-12-16T17:51:09
null
null
Journal-ref: Reinforcement Learning Journal, Volume 5, 2024, Pages 2359-2372
Published at Reinforcement Learning Conference (RLC) 2024. Website: http://lfzhao.com/map-nav/
https://arxiv.org/abs/2412.12024v1
https://arxiv.org/pdf/2412.12024v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.09582v2
2412.09582
2
Neptune: The Long Orbit to Benchmarking Long Video Understanding
We introduce Neptune, a benchmark for long video understanding that requires reasoning over long time horizons and across different modalities. Many existing video datasets and models are focused on short clips (10s-30s). While some long video datasets do exist, they can often be solved by powerful image models applied...
212
[ "Arsha Nagrani", "Mingda Zhang", "Ramin Mehran", "Rachel Hornung", "Nitesh Bharadwaj Gundavarapu", "Nilpa Jha", "Austin Myers", "Xingyi Zhou", "Boqing Gong", "Cordelia Schmid", "Mikhail Sirotenko", "Yukun Zhu", "Tobias Weyand" ]
13
Arsha Nagrani
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CV" ]
true
2024-12-12T18:54:48
2024-12-12T00:00:00
2025-01-18T00:52:42
null
null
null
null
https://arxiv.org/abs/2412.09582v2
https://arxiv.org/pdf/2412.09582v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.12147v1
2412.12147
1
Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control
Generalizing across robot embodiments and tasks is crucial for adaptive robotic systems. Modular policy learning approaches adapt to new embodiments but are limited to specific tasks, while few-shot imitation learning (IL) approaches often focus on a single embodiment. In this paper, we introduce a few-shot behavior cl...
153
[ "Seongwoong Cho", "Donggyun Kim", "Jinwoo Lee", "Seunghoon Hong" ]
4
Seongwoong Cho
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.RO" ]
true
2024-12-10T19:10:17
2024-12-10T00:00:00
2024-12-10T19:10:17
null
null
null
NeurIPS 2024
https://arxiv.org/abs/2412.12147v1
https://arxiv.org/pdf/2412.12147v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2412.06771v3
2412.06771
3
Proactive Agents for Multi-Turn Text-to-Image Generation Under Uncertainty
User prompts for generative AI models are often underspecified, leading to a misalignment between the user intent and models' understanding. As a result, users commonly have to painstakingly refine their prompts. We study this alignment problem in text-to-image (T2I) generation and propose a prototype for proactive T2I...
226
[ "Meera Hahn", "Wenjun Zeng", "Nithish Kannen", "Rich Galt", "Kartikeya Badola", "Been Kim", "Zi Wang" ]
7
Meera Hahn
[]
cs.AI
[ "cs.AI", "cs.CV", "cs.LG" ]
true
2024-12-09T18:56:32
2024-12-09T00:00:00
2025-10-24T02:32:07
null
null
International Conference on Machine Learning, 2025
null
https://arxiv.org/abs/2412.06771v3
https://arxiv.org/pdf/2412.06771v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2411.07038v1
2411.07038
1
Designing Reliable Experiments with Generative Agent-Based Modeling: A Comprehensive Guide Using Concordia by Google DeepMind
In social sciences, researchers often face challenges when conducting large-scale experiments, particularly due to the simulations' complexity and the lack of technical expertise required to develop such frameworks. Agent-Based Modeling (ABM) is a computational approach that simulates agents' actions and interactions t...
126
[ "Alejandro Leonardo García Navarro", "Nataliia Koneva", "Alfonso Sánchez-Macián", "José Alberto Hernández", "Manuel Goyanes" ]
5
Alejandro Leonardo García Navarro
[]
cs.AI
[ "cs.AI" ]
false
2024-11-11T14:45:08
2024-11-11T00:00:00
2024-11-11T14:45:08
null
null
null
null
https://arxiv.org/abs/2411.07038v1
https://arxiv.org/pdf/2411.07038v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2411.06174v1
2411.06174
1
State Chrono Representation for Enhancing Generalization in Reinforcement Learning
In reinforcement learning with image-based inputs, it is crucial to establish a robust and generalizable state representation. Recent advancements in metric learning, such as deep bisimulation metric approaches, have shown promising results in learning structured low-dimensional representation space from pixel observat...
189
[ "Jianda Chen", "Wen Zheng Terence Ng", "Zichen Chen", "Sinno Jialin Pan", "Tianwei Zhang" ]
5
Jianda Chen
[]
cs.LG
[ "cs.LG", "cs.RO" ]
true
2024-11-09T13:12:34
2024-11-09T00:00:00
2024-11-09T13:12:34
null
null
38th Conference on Neural Information Processing Systems (NeurIPS 2024)
null
https://arxiv.org/abs/2411.06174v1
https://arxiv.org/pdf/2411.06174v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2411.03604v1
2411.03604
1
Temporal-Difference Learning Using Distributed Error Signals
A computational problem in biological reward-based learning is how credit assignment is performed in the nucleus accumbens (NAc). Much research suggests that NAc dopamine encodes temporal-difference (TD) errors for learning value predictions. However, dopamine is synchronously distributed in regionally homogeneous conc...
130
[ "Jonas Guan", "Shon Eduard Verch", "Claas Voelcker", "Ethan C. Jackson", "Nicolas Papernot", "William A. Cunningham" ]
6
Jonas Guan
[]
cs.LG
[ "cs.LG", "cs.NE" ]
true
2024-11-06T01:49:13
2024-11-06T00:00:00
2024-11-06T01:49:13
null
null
null
10 pages, to be published at NeurIPS 2024
https://arxiv.org/abs/2411.03604v1
https://arxiv.org/pdf/2411.03604v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.21939v3
2410.21939
3
HonestCyberEval: An AI Cyber Risk Benchmark for Automated Software Exploitation
We introduce HonestCyberEval, a new benchmark for assessing AI models' capabilities and risks in automated software exploitation, focusing on their ability to detect and exploit vulnerabilities in real-world software systems. Our evaluation leverages the Nginx web server repository augmented with synthetic vulnerabilit...
118
[ "Dan Ristea", "Vasilios Mavroudis" ]
2
Dan Ristea
[]
cs.CR
[ "cs.CR", "cs.AI" ]
true
2024-10-29T10:57:11
2024-10-29T00:00:00
2025-08-26T09:59:15
null
null
null
null
https://arxiv.org/abs/2410.21939v3
https://arxiv.org/pdf/2410.21939v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.17428v3
2410.17428
3
Uncovering RL Integration in SSL Loss: Objective-Specific Implications for Data-Efficient RL
In this study, we investigate the effect of SSL objective modifications within the SPR framework, focusing on specific adjustments such as terminal state masking and prioritized replay weighting, which were not explicitly addressed in the original design. While these modifications are specific to RL, they are not unive...
157
[ "Ömer Veysel Çağatan", "Barış Akgün" ]
2
Ömer Veysel Çağatan
[]
cs.LG
[ "cs.LG" ]
false
2024-10-22T20:57:10
2024-10-22T00:00:00
2025-07-10T11:21:38
null
null
null
RLC 2025, Neurips SSL Workshop 2024
https://arxiv.org/abs/2410.17428v3
https://arxiv.org/pdf/2410.17428v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.16790v2
2410.16790
2
Curriculum Reinforcement Learning for Complex Reward Functions
Reinforcement learning (RL) has emerged as a powerful tool for tackling control problems, but its practical application is often hindered by the complexity arising from intricate reward functions with multiple terms. The reward hypothesis posits that any objective can be encapsulated in a scalar reward function, yet ba...
249
[ "Kilian Freitag", "Kristian Ceder", "Rita Laezza", "Knut Åkesson", "Morteza Haghir Chehreghani" ]
5
Kilian Freitag
[]
cs.LG
[ "cs.LG", "cs.RO" ]
true
2024-10-22T08:07:44
2024-10-22T00:00:00
2025-02-10T10:42:49
null
null
null
null
https://arxiv.org/abs/2410.16790v2
https://arxiv.org/pdf/2410.16790v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.16512v2
2410.16512
2
TIPS: Text-Image Pretraining with Spatial awareness
While image-text representation learning has become very popular in recent years, existing models tend to lack spatial awareness and have limited direct applicability for dense understanding tasks. For this reason, self-supervised image-only pretraining is still the go-to method for many dense vision applications (e.g....
245
[ "Kevis-Kokitsi Maninis", "Kaifeng Chen", "Soham Ghosh", "Arjun Karpur", "Koert Chen", "Ye Xia", "Bingyi Cao", "Daniel Salz", "Guangxing Han", "Jan Dlabal", "Dan Gnanapragasam", "Mojtaba Seyedhosseini", "Howard Zhou", "Andre Araujo" ]
14
Kevis-Kokitsi Maninis
[]
cs.CV
[ "cs.CV" ]
false
2024-10-21T21:05:04
2024-10-21T00:00:00
2025-03-07T19:38:42
null
null
null
ICLR2025 camera-ready + appendix
https://arxiv.org/abs/2410.16512v2
https://arxiv.org/pdf/2410.16512v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.09972v1
2410.09972
1
Make the Pertinent Salient: Task-Relevant Reconstruction for Visual Control with Distractions
Recent advancements in Model-Based Reinforcement Learning (MBRL) have made it a powerful tool for visual control tasks. Despite improved data efficiency, it remains challenging to train MBRL agents with generalizable perception. Training in the presence of visual distractions is particularly difficult due to the high v...
231
[ "Kyungmin Kim", "JB Lanier", "Pierre Baldi", "Charless Fowlkes", "Roy Fox" ]
5
Kyungmin Kim
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CV", "cs.RO" ]
true
2024-10-13T19:24:07
2024-10-13T00:00:00
2024-10-13T19:24:07
null
null
null
null
https://arxiv.org/abs/2410.09972v1
https://arxiv.org/pdf/2410.09972v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.08896v2
2410.08896
2
MAD-TD: Model-Augmented Data stabilizes High Update Ratio RL
Building deep reinforcement learning (RL) agents that find a good policy with few samples has proven notoriously challenging. To achieve sample efficiency, recent work has explored updating neural networks with large numbers of gradient steps for every new sample. While such high update-to-data (UTD) ratios have shown ...
204
[ "Claas A Voelcker", "Marcel Hussing", "Eric Eaton", "Amir-massoud Farahmand", "Igor Gilitschenski" ]
5
Claas A Voelcker
[]
cs.LG
[ "cs.LG" ]
false
2024-10-11T15:13:17
2024-10-11T00:00:00
2025-04-02T18:17:24
null
null
null
null
https://arxiv.org/abs/2410.08896v2
https://arxiv.org/pdf/2410.08896v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.07994v3
2410.07994
3
Neuroplastic Expansion in Deep Reinforcement Learning
The loss of plasticity in learning agents, analogous to the solidification of neural pathways in biological brains, significantly impedes learning and adaptation in reinforcement learning due to its non-stationary nature. To address this fundamental challenge, we propose a novel approach, {\it Neuroplastic Expansion} (...
167
[ "Jiashun Liu", "Johan Obando-Ceron", "Aaron Courville", "Ling Pan" ]
4
Jiashun Liu
[]
cs.LG
[ "cs.LG" ]
false
2024-10-10T14:51:14
2024-10-10T00:00:00
2025-06-02T12:04:04
null
null
null
null
https://arxiv.org/abs/2410.07994v3
https://arxiv.org/pdf/2410.07994v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.07836v6
2410.07836
6
Masked Generative Priors Improve World Models Sequence Modelling Capabilities
Deep Reinforcement Learning (RL) has become the leading approach for creating artificial agents in complex environments. Model-based approaches, which are RL methods with world models that predict environment dynamics, are among the most promising directions for improving data efficiency, forming a critical step toward...
244
[ "Cristian Meo", "Mircea Lica", "Zarif Ikram", "Akihiro Nakano", "Vedant Shah", "Aniket Rajiv Didolkar", "Dianbo Liu", "Anirudh Goyal", "Justin Dauwels" ]
9
Cristian Meo
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-10-10T11:52:07
2024-10-10T00:00:00
2025-10-22T15:16:35
null
null
null
null
https://arxiv.org/abs/2410.07836v6
https://arxiv.org/pdf/2410.07836v6
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.06621v1
2410.06621
1
Effective Exploration Based on the Structural Information Principles
Traditional information theory provides a valuable foundation for Reinforcement Learning, particularly through representation learning and entropy maximization for agent exploration. However, existing methods primarily concentrate on modeling the uncertainty associated with RL's random variables, neglecting the inheren...
193
[ "Xianghua Zeng", "Hao Peng", "Angsheng Li" ]
3
Xianghua Zeng
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-10-09T07:19:16
2024-10-09T00:00:00
2024-10-09T07:19:16
null
null
null
10 pages in main paper and 15 pages in appendix
https://arxiv.org/abs/2410.06621v1
https://arxiv.org/pdf/2410.06621v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.06317v2
2410.06317
2
Learning in complex action spaces without policy gradients
While conventional wisdom holds that policy gradient methods are better suited to complex action spaces than action-value methods, foundational work has shown that the two paradigms are equivalent in small, finite action spaces (O'Donoghue et al., 2017; Schulman et al., 2017a). This raises the question of why their com...
179
[ "Arash Tavakoli", "Sina Ghiassian", "Nemanja Rakićević" ]
3
Arash Tavakoli
[]
cs.LG
[ "cs.LG", "cs.AI", "stat.ML" ]
true
2024-10-08T19:49:34
2024-10-08T00:00:00
2025-09-01T01:26:06
null
null
Transactions on Machine Learning Research (2025)
Published in TMLR (2025). Code: https://github.com/atavakol/qmle
https://arxiv.org/abs/2410.06317v2
https://arxiv.org/pdf/2410.06317v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.04553v1
2410.04553
1
Bisimulation metric for Model Predictive Control
Model-based reinforcement learning has shown promise for improving sample efficiency and decision-making in complex environments. However, existing methods face challenges in training stability, robustness to noise, and computational efficiency. In this paper, we propose Bisimulation Metric for Model Predictive Control...
127
[ "Yutaka Shimizu", "Masayoshi Tomizuka" ]
2
Yutaka Shimizu
[]
cs.LG
[ "cs.LG", "eess.SY" ]
true
2024-10-06T17:12:10
2024-10-06T00:00:00
2024-10-06T17:12:10
null
null
null
null
https://arxiv.org/abs/2410.04553v1
https://arxiv.org/pdf/2410.04553v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2410.00051v2
2410.00051
2
Generalizing Consistency Policy to Visual RL with Prioritized Proximal Experience Regularization
With high-dimensional state spaces, visual reinforcement learning (RL) faces significant challenges in exploitation and exploration, resulting in low sample efficiency and training stability. As a time-efficient diffusion model, although consistency models have been validated in online state-based RL, it is still an op...
166
[ "Haoran Li", "Zhennan Jiang", "Yuhui Chen", "Dongbin Zhao" ]
4
Haoran Li
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CV" ]
true
2024-09-28T09:24:10
2024-09-28T00:00:00
2024-10-29T09:44:15
null
null
null
Accepted at the Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS2024)
https://arxiv.org/abs/2410.00051v2
https://arxiv.org/pdf/2410.00051v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2409.18330v1
2409.18330
1
DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors
Learning from previously collected data via behavioral cloning or offline reinforcement learning (RL) is a powerful recipe for scaling generalist agents by avoiding the need for expensive online learning. Despite strong generalization in some respects, agents are often remarkably brittle to minor visual variations in c...
242
[ "Joseph Ortiz", "Antoine Dedieu", "Wolfgang Lehrach", "Swaroop Guntupalli", "Carter Wendelken", "Ahmad Humayun", "Guangyao Zhou", "Sivaramakrishnan Swaminathan", "Miguel Lázaro-Gredilla", "Kevin Murphy" ]
10
Joseph Ortiz
[]
cs.LG
[ "cs.LG" ]
false
2024-09-26T23:07:01
2024-09-26T00:00:00
2024-09-26T23:07:01
null
null
null
NeurIPS 2024 Datasets and Benchmarks Track. Dataset available at: https://github.com/google-deepmind/dmc_vision_benchmark
https://arxiv.org/abs/2409.18330v1
https://arxiv.org/pdf/2409.18330v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2409.08751v1
2409.08751
1
A Grading Rubric for AI Safety Frameworks
Over the past year, artificial intelligence (AI) companies have been increasingly adopting AI safety frameworks. These frameworks outline how companies intend to keep the potential risks associated with developing and deploying frontier AI systems to an acceptable level. Major players like Anthropic, OpenAI, and Google...
181
[ "Jide Alaga", "Jonas Schuett", "Markus Anderljung" ]
3
Jide Alaga
[]
cs.CY
[ "cs.CY" ]
false
2024-09-13T12:01:55
2024-09-13T00:00:00
2024-09-13T12:01:55
null
null
null
16 pages, 4 tables
https://arxiv.org/abs/2409.08751v1
https://arxiv.org/pdf/2409.08751v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2409.07878v2
2409.07878
2
Mapping Technical Safety Research at AI Companies: A literature review and incentives analysis
As AI systems become more advanced, concerns about large-scale risks from misuse or accidents have grown. This report analyzes the technical research into safe AI development being conducted by three leading AI companies: Anthropic, Google DeepMind, and OpenAI. We define safe AI development as developing AI systems tha...
282
[ "Oscar Delaney", "Oliver Guest", "Zoe Williams" ]
3
Oscar Delaney
[]
cs.CY
[ "cs.CY" ]
false
2024-09-12T09:34:55
2024-09-12T00:00:00
2024-09-25T16:19:25
null
null
null
Sep-25 update adds papers that we missed in the original dataset, as well as more acknowledgements
https://arxiv.org/abs/2409.07878v2
https://arxiv.org/pdf/2409.07878v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2409.07268v2
2409.07268
2
Multi-Type Preference Learning: Empowering Preference-Based Reinforcement Learning with Equal Preferences
Preference-Based reinforcement learning (PBRL) learns directly from the preferences of human teachers regarding agent behaviors without needing meticulously designed reward functions. However, existing PBRL methods often learn primarily from explicit preferences, neglecting the possibility that teachers may choose equa...
183
[ "Ziang Liu", "Junjie Xu", "Xingjiao Wu", "Jing Yang", "Liang He" ]
5
Ziang Liu
[]
cs.LG
[ "cs.LG" ]
false
2024-09-11T13:43:49
2024-09-11T00:00:00
2024-10-15T06:32:25
null
null
null
8 pages, 6 figures
https://arxiv.org/abs/2409.07268v2
https://arxiv.org/pdf/2409.07268v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.14587v2
2408.14587
2
Efficient fine-tuning of 37-level GraphCast with the Canadian global deterministic analysis
This work describes a process for efficiently fine-tuning the GraphCast data-driven forecast model to simulate another analysis system, here the Global Deterministic Prediction System (GDPS) of Environment and Climate Change Canada (ECCC). Using two years of training data (July 2019 -- December 2021) and 37 GPU-days of...
152
[ "Christopher Subich" ]
1
Christopher Subich
[]
cs.LG
[ "cs.LG", "physics.ao-ph" ]
true
2024-08-26T19:16:08
2024-08-26T00:00:00
2025-04-25T13:46:36
10.1175/AIES-D-24-0101.1
https://doi.org/10.1175/AIES-D-24-0101.1
Artificial Intelligence for the Earth Systems 4(3) Jul. 2025
null
https://arxiv.org/abs/2408.14587v2
https://arxiv.org/pdf/2408.14587v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.11527v3
2408.11527
3
The Vizier Gaussian Process Bandit Algorithm
Google Vizier has performed millions of optimizations and accelerated numerous research and production systems at Google, demonstrating the success of Bayesian optimization as a large-scale service. Over multiple years, its algorithm has been improved considerably, through the collective experiences of numerous researc...
86
[ "Xingyou Song", "Qiuyi Zhang", "Chansoo Lee", "Emily Fertig", "Tzu-Kuo Huang", "Lior Belenki", "Greg Kochanski", "Setareh Ariafar", "Srinivas Vasudevan", "Sagi Perel", "Daniel Golovin" ]
11
Xingyou Song
[]
cs.LG
[ "cs.LG", "cs.AI", "math.OC" ]
true
2024-08-21T11:06:02
2024-08-21T00:00:00
2024-12-06T17:31:39
null
null
null
Google DeepMind Technical Report. Code can be found in https://github.com/google/vizier
https://arxiv.org/abs/2408.11527v3
https://arxiv.org/pdf/2408.11527v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.06101v1
2408.06101
1
Generalization capabilities of MeshGraphNets to unseen geometries for fluid dynamics
This works investigates the generalization capabilities of MeshGraphNets (MGN) [Pfaff et al. Learning Mesh-Based Simulation with Graph Networks. ICML 2021] to unseen geometries for fluid dynamics, e.g. predicting the flow around a new obstacle that was not part of the training data. For this purpose, we create a new be...
132
[ "Robin Schmöcker", "Alexander Henkes", "Julian Roth", "Thomas Wick" ]
4
Robin Schmöcker
[]
cs.LG
[ "cs.LG", "cs.AI", "math.NA", "physics.flu-dyn" ]
true
2024-08-12T12:32:15
2024-08-12T00:00:00
2024-08-12T12:32:15
null
null
null
null
https://arxiv.org/abs/2408.06101v1
https://arxiv.org/pdf/2408.06101v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.05781v2
2408.05781
2
CURLing the Dream: Contrastive Representations for World Modeling in Reinforcement Learning
In this work, we present Curled-Dreamer, a novel reinforcement learning algorithm that integrates contrastive learning into the DreamerV3 framework to enhance performance in visual reinforcement learning tasks. By incorporating the contrastive loss from the CURL algorithm and a reconstruction loss from autoencoder, Cur...
112
[ "Victor Augusto Kich", "Jair Augusto Bottega", "Raul Steinmetz", "Ricardo Bedin Grando", "Ayano Yorozu", "Akihisa Ohya" ]
6
Victor Augusto Kich
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CV" ]
true
2024-08-11T14:13:22
2024-08-11T00:00:00
2024-08-31T21:20:16
null
null
null
Paper accepted for 24th International Conference on Control, Automation and Systems (ICCAS)
https://arxiv.org/abs/2408.05781v2
https://arxiv.org/pdf/2408.05781v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.04841v3
2408.04841
3
Kolmogorov-Arnold Network for Online Reinforcement Learning
Kolmogorov-Arnold Networks (KANs) have shown potential as an alternative to Multi-Layer Perceptrons (MLPs) in neural networks, providing universal function approximation with fewer parameters and reduced memory usage. In this paper, we explore the use of KANs as function approximators within the Proximal Policy Optimiz...
102
[ "Victor Augusto Kich", "Jair Augusto Bottega", "Raul Steinmetz", "Ricardo Bedin Grando", "Ayano Yorozu", "Akihisa Ohya" ]
6
Victor Augusto Kich
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-08-09T03:32:37
2024-08-09T00:00:00
2024-08-31T21:01:06
null
null
null
Paper accepted at 24th International Conference on Control, Automation and Systems (ICCAS)
https://arxiv.org/abs/2408.04841v3
https://arxiv.org/pdf/2408.04841v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2408.00521v2
2408.00521
2
A new approach for encoding code and assisting code understanding
Some companies (e.g., Microsoft Research and Google DeepMind) have discovered some of the limitations of GPTs' autoregressive paradigm next-word prediction, manifested in the model's lack of planning, working memory, backtracking, and reasoning skills. GPTs rely on a local and greedy process of generating the next word...
282
[ "Mengdan Fan", "Wei Zhang", "Haiyan Zhao", "Zhi Jin" ]
4
Mengdan Fan
[]
cs.AI
[ "cs.AI" ]
false
2024-08-01T12:52:48
2024-08-01T00:00:00
2025-03-23T11:30:23
null
null
null
10 page, 14 figures
https://arxiv.org/abs/2408.00521v2
https://arxiv.org/pdf/2408.00521v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2407.19447v1
2407.19447
1
Nudging Consent and the New Opt Out System to the Processing of Health Data in England
This chapter examines the challenges of the revised opt out system and the secondary use of health data in England. The analysis of this data could be very valuable for science and medical treatment as well as for the discovery of new drugs. For this reason, the UK government established the care.data program in 2013. ...
298
[ "Janos Meszaros", "Chih-hsing Ho", "Marcelo Corrales Compagnucci" ]
3
Janos Meszaros
[]
cs.CY
[ "cs.CY", "cs.AI" ]
true
2024-07-28T09:49:37
2024-07-28T00:00:00
2024-07-28T09:49:37
null
null
null
null
https://arxiv.org/abs/2407.19447v1
https://arxiv.org/pdf/2407.19447v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2407.04315v1
2407.04315
1
Gradient-based Regularization for Action Smoothness in Robotic Control with Reinforcement Learning
Deep Reinforcement Learning (DRL) has achieved remarkable success, ranging from complex computer games to real-world applications, showing the potential for intelligent agents capable of learning in dynamic environments. However, its application in real-world scenarios presents challenges, including the jerky problem, ...
197
[ "I Lee", "Hoang-Giang Cao", "Cong-Tinh Dao", "Yu-Cheng Chen", "I-Chen Wu" ]
5
I Lee
[]
cs.RO
[ "cs.RO" ]
false
2024-07-05T07:35:32
2024-07-05T00:00:00
2024-07-05T07:35:32
null
null
null
Accepted to IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2024
https://arxiv.org/abs/2407.04315v1
https://arxiv.org/pdf/2407.04315v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2407.12034v2
2407.12034
2
Understanding Transformers via N-gram Statistics
Transformer based large-language models (LLMs) display extreme proficiency with language yet a precise understanding of how they work remains elusive. One way of demystifying transformer predictions would be to describe how they depend on their context in terms of simple template functions. This paper takes a first ste...
184
[ "Timothy Nguyen" ]
1
Timothy Nguyen
[]
cs.CL
[ "cs.CL", "cs.AI", "cs.LG" ]
true
2024-06-30T22:18:49
2024-06-30T00:00:00
2024-11-05T10:24:42
null
null
null
NeurIPS 2024. Datasets and N-gram statistics open-sourced: https://github.com/google-deepmind/transformer_ngrams
https://arxiv.org/abs/2407.12034v2
https://arxiv.org/pdf/2407.12034v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2407.00633v2
2407.00633
2
DEAR: Disentangled Environment and Agent Representations for Reinforcement Learning without Reconstruction
Reinforcement Learning (RL) algorithms can learn robotic control tasks from visual observations, but they often require a large amount of data, especially when the visual scene is complex and unstructured. In this paper, we explore how the agent's knowledge of its shape can improve the sample efficiency of visual RL me...
178
[ "Ameya Pore", "Riccardo Muradore", "Diego Dall'Alba" ]
3
Ameya Pore
[]
cs.CV
[ "cs.CV" ]
false
2024-06-30T09:15:21
2024-06-30T00:00:00
2024-10-22T18:52:19
null
null
null
6 pages, 7 figures, 2 tables. Accepted at 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2024)
https://arxiv.org/abs/2407.00633v2
https://arxiv.org/pdf/2407.00633v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.16255v2
2406.16255
2
Uncertainty-Aware Reward-Free Exploration with General Function Approximation
Mastering multiple tasks through exploration and learning in an environment poses a significant challenge in reinforcement learning (RL). Unsupervised RL has been introduced to address this challenge by training policies with intrinsic rewards rather than extrinsic rewards. However, current intrinsic reward designs and...
194
[ "Junkai Zhang", "Weitong Zhang", "Dongruo Zhou", "Quanquan Gu" ]
4
Junkai Zhang
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-06-24T01:37:18
2024-06-24T00:00:00
2024-06-30T03:55:48
null
null
null
32 pages, 5 figures, 4 tables, accepted by ICML 2024
https://arxiv.org/abs/2406.16255v2
https://arxiv.org/pdf/2406.16255v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.13121v1
2406.13121
1
Can Long-Context Language Models Subsume Retrieval, RAG, SQL, and More?
Long-context language models (LCLMs) have the potential to revolutionize our approach to tasks traditionally reliant on external tools like retrieval systems or databases. Leveraging LCLMs' ability to natively ingest and process entire corpora of information offers numerous advantages. It enhances user-friendliness by ...
182
[ "Jinhyuk Lee", "Anthony Chen", "Zhuyun Dai", "Dheeru Dua", "Devendra Singh Sachan", "Michael Boratko", "Yi Luan", "Sébastien M. R. Arnold", "Vincent Perot", "Siddharth Dalmia", "Hexiang Hu", "Xudong Lin", "Panupong Pasupat", "Aida Amini", "Jeremy R. Cole", "Sebastian Riedel", "Iftekh...
19
Jinhyuk Lee
[]
cs.CL
[ "cs.CL", "cs.AI", "cs.IR" ]
true
2024-06-19T00:28:58
2024-06-19T00:00:00
2024-06-19T00:28:58
null
null
null
29 pages. Dataset available at https://github.com/google-deepmind/loft
https://arxiv.org/abs/2406.13121v1
https://arxiv.org/pdf/2406.13121v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.09159v1
2406.09159
1
ALPHAGMUT: A Rationale-Guided Alpha Shape Graph Neural Network to Evaluate Mutation Effects
In silico methods evaluating the mutation effects of missense mutations are providing an important approach for understanding mutations in personal genomes and identifying disease-relevant biomarkers. However, existing methods, including deep learning methods, heavily rely on sequence-aware information, and do not full...
205
[ "Boshen Wang", "Bowei Ye", "Lin Xu", "Jie Liang" ]
4
Boshen Wang
[]
q-bio.QM
[ "q-bio.QM", "cs.AI", "cs.CG", "q-bio.GN" ]
true
2024-06-13T14:22:12
2024-06-13T00:00:00
2024-06-13T14:22:12
null
null
null
2 figures, 2 tables
https://arxiv.org/abs/2406.09159v1
https://arxiv.org/pdf/2406.09159v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2407.10981v1
2407.10981
1
Systematic Literature Review of AI-enabled Spectrum Management in 6G and Future Networks
Artificial Intelligence (AI) has advanced significantly in various domains like healthcare, finance, and cybersecurity, with successes such as DeepMind's medical imaging and Tesla's autonomous vehicles. As telecommunications transition from 5G to 6G, integrating AI is crucial for complex demands like data processing, n...
221
[ "Bushra Sabir", "Shuiqiao Yang", "David Nguyen", "Nan Wu", "Alsharif Abuadbba", "Hajime Suzuki", "Shangqi Lai", "Wei Ni", "Ding Ming", "Surya Nepal" ]
10
Bushra Sabir
[]
cs.NI
[ "cs.NI", "cs.CR" ]
true
2024-06-12T11:31:42
2024-06-12T00:00:00
2024-06-12T11:31:42
null
null
null
35 pages
https://arxiv.org/abs/2407.10981v1
https://arxiv.org/pdf/2407.10981v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.07539v2
2406.07539
2
BAKU: An Efficient Transformer for Multi-Task Policy Learning
Training generalist agents capable of solving diverse tasks is challenging, often requiring large datasets of expert demonstrations. This is particularly problematic in robotics, where each data point requires physical execution of actions in the real world. Thus, there is a pressing need for architectures that can eff...
160
[ "Siddhant Haldar", "Zhuoran Peng", "Lerrel Pinto" ]
3
Siddhant Haldar
[]
cs.RO
[ "cs.RO" ]
false
2024-06-11T17:58:54
2024-06-11T00:00:00
2024-07-16T20:20:07
null
null
null
null
https://arxiv.org/abs/2406.07539v2
https://arxiv.org/pdf/2406.07539v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.05953v1
2406.05953
1
Decoupling regularization from the action space
Regularized reinforcement learning (RL), particularly the entropy-regularized kind, has gained traction in optimal control and inverse RL. While standard unregularized RL methods remain unaffected by changes in the number of actions, we show that it can severely impact their regularized counterparts. This paper demonst...
154
[ "Sobhan Mohammadpour", "Emma Frejinger", "Pierre-Luc Bacon" ]
3
Sobhan Mohammadpour
[]
cs.LG
[ "cs.LG" ]
false
2024-06-10T01:20:31
2024-06-10T00:00:00
2024-06-10T01:20:31
null
null
The Twelfth International Conference on Learning Representations, 2024
null
https://arxiv.org/abs/2406.05953v1
https://arxiv.org/pdf/2406.05953v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.04329v4
2406.04329
4
Simplified and Generalized Masked Diffusion for Discrete Data
Masked (or absorbing) diffusion is actively explored as an alternative to autoregressive models for generative modeling of discrete data. However, existing work in this area has been hindered by unnecessarily complex model formulations and unclear relationships between different perspectives, leading to suboptimal para...
174
[ "Jiaxin Shi", "Kehang Han", "Zhe Wang", "Arnaud Doucet", "Michalis K. Titsias" ]
5
Jiaxin Shi
[]
cs.LG
[ "cs.LG", "stat.ML" ]
true
2024-06-06T17:59:10
2024-06-06T00:00:00
2025-01-16T08:46:16
null
null
null
NeurIPS 2024. Code is available at: https://github.com/google-deepmind/md4
https://arxiv.org/abs/2406.04329v4
https://arxiv.org/pdf/2406.04329v4
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.04229v1
2406.04229
1
The CLRS-Text Algorithmic Reasoning Language Benchmark
Eliciting reasoning capabilities from language models (LMs) is a critical direction on the path towards building intelligent systems. Most recent studies dedicated to reasoning focus on out-of-distribution performance on procedurally-generated synthetic benchmarks, bespoke-built to evaluate specific skills only. This t...
178
[ "Larisa Markeeva", "Sean McLeish", "Borja Ibarz", "Wilfried Bounsi", "Olga Kozlova", "Alex Vitvitskyi", "Charles Blundell", "Tom Goldstein", "Avi Schwarzschild", "Petar Veličković" ]
10
Larisa Markeeva
[]
cs.LG
[ "cs.LG", "cs.AI", "cs.CL", "cs.DS", "stat.ML" ]
true
2024-06-06T16:29:25
2024-06-06T00:00:00
2024-06-06T16:29:25
null
null
null
Preprint, under review. Comments welcome
https://arxiv.org/abs/2406.04229v1
https://arxiv.org/pdf/2406.04229v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.02696v1
2406.02696
1
iQRL -- Implicitly Quantized Representations for Sample-efficient Reinforcement Learning
Learning representations for reinforcement learning (RL) has shown much promise for continuous control. We propose an efficient representation learning method using only a self-supervised latent-state consistency loss. Our approach employs an encoder and a dynamics model to map observations to latent states and predict...
107
[ "Aidan Scannell", "Kalle Kujanpää", "Yi Zhao", "Mohammadreza Nakhaei", "Arno Solin", "Joni Pajarinen" ]
6
Aidan Scannell
[]
cs.LG
[ "cs.LG" ]
false
2024-06-04T18:15:44
2024-06-04T00:00:00
2024-06-04T18:15:44
null
null
null
9 pages, 11 figures
https://arxiv.org/abs/2406.02696v1
https://arxiv.org/pdf/2406.02696v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2406.01423v4
2406.01423
4
Value Improved Actor Critic Algorithms
To learn approximately optimal acting policies for decision problems, modern Actor Critic algorithms rely on deep Neural Networks (DNNs) to parameterize the acting policy and greedification operators to iteratively improve it. The reliance on DNNs suggests an improvement that is gradient based, which is per step much l...
195
[ "Yaniv Oren", "Moritz A. Zanger", "Pascal R. van der Vaart", "Mustafa Mert Celikok", "Matthijs T. J. Spaan", "Wendelin Bohmer" ]
6
Yaniv Oren
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-06-03T15:24:15
2024-06-03T00:00:00
2026-01-16T12:08:53
null
null
null
null
https://arxiv.org/abs/2406.01423v4
https://arxiv.org/pdf/2406.01423v4
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2405.16158v3
2405.16158
3
Bigger, Regularized, Optimistic: scaling for compute and sample-efficient continuous control
Sample efficiency in Reinforcement Learning (RL) has traditionally been driven by algorithmic enhancements. In this work, we demonstrate that scaling can also lead to substantial improvements. We conduct a thorough investigation into the interplay of scaling model capacity and domain-specific RL enhancements. These emp...
125
[ "Michal Nauman", "Mateusz Ostaszewski", "Krzysztof Jankowski", "Piotr Miłoś", "Marek Cygan" ]
5
Michal Nauman
[]
cs.LG
[ "cs.LG" ]
false
2024-05-25T09:53:25
2024-05-25T00:00:00
2024-12-03T08:42:49
null
null
null
NeurIPS 2024 Spotlight
https://arxiv.org/abs/2405.16158v3
https://arxiv.org/pdf/2405.16158v3
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2405.15083v1
2405.15083
1
MuDreamer: Learning Predictive World Models without Reconstruction
The DreamerV3 agent recently demonstrated state-of-the-art performance in diverse domains, learning powerful world models in latent space using a pixel reconstruction loss. However, while the reconstruction loss is essential to Dreamer's performance, it also necessitates modeling unnecessary information. Consequently, ...
203
[ "Maxime Burchi", "Radu Timofte" ]
2
Maxime Burchi
[]
cs.AI
[ "cs.AI", "cs.CV" ]
true
2024-05-23T22:09:01
2024-05-23T00:00:00
2024-05-23T22:09:01
null
null
null
null
https://arxiv.org/abs/2405.15083v1
https://arxiv.org/pdf/2405.15083v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2405.13960v1
2405.13960
1
Learning To Play Atari Games Using Dueling Q-Learning and Hebbian Plasticity
In this work, an advanced deep reinforcement learning architecture is used to train neural network agents playing atari games. Given only the raw game pixels, action space, and reward information, the system can train agents to play any Atari game. At first, this system uses advanced techniques like deep Q-networks and...
145
[ "Md Ashfaq Salehin" ]
1
Md Ashfaq Salehin
[]
cs.AI
[ "cs.AI", "cs.LG" ]
true
2024-05-22T19:55:33
2024-05-22T00:00:00
2024-05-22T19:55:33
null
null
null
null
https://arxiv.org/abs/2405.13960v1
https://arxiv.org/pdf/2405.13960v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2405.09788v2
2405.09788
2
Synthesizing Proteins on the Graphics Card. Protein Folding and the Limits of Critical AI Studies
This paper investigates the application of the transformer architecture in protein folding, as exemplified by DeepMind's AlphaFold project, and its implications for the understanding of so-called large language models. The prevailing discourse often assumes a ready-made analogy between proteins, encoded as sequences of...
250
[ "Fabian Offert", "Paul Kim", "Qiaoyu Cai" ]
3
Fabian Offert
[]
cs.CY
[ "cs.CY" ]
false
2024-05-16T03:24:05
2024-05-16T00:00:00
2024-12-07T18:24:44
null
null
null
null
https://arxiv.org/abs/2405.09788v2
https://arxiv.org/pdf/2405.09788v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2404.18896v2
2404.18896
2
Overcoming Knowledge Barriers: Online Imitation Learning from Visual Observation with Pretrained World Models
Pretraining and finetuning models has become increasingly popular in decision-making. But there are still serious impediments in Imitation Learning from Observation (ILfO) with pretrained models. This study identifies two primary obstacles: the Embodiment Knowledge Barrier (EKB) and the Demonstration Knowledge Barrier ...
172
[ "Xingyuan Zhang", "Philip Becker-Ehmck", "Patrick van der Smagt", "Maximilian Karl" ]
4
Xingyuan Zhang
[]
cs.LG
[ "cs.LG" ]
false
2024-04-29T17:33:52
2024-04-29T00:00:00
2025-04-23T19:11:30
null
null
null
Accepted at TMLR
https://arxiv.org/abs/2404.18896v2
https://arxiv.org/pdf/2404.18896v2
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2404.14068v1
2404.14068
1
Holistic Safety and Responsibility Evaluations of Advanced AI Models
Safety and responsibility evaluations of advanced AI models are a critical but developing field of research and practice. In the development of Google DeepMind's advanced AI models, we innovated on and applied a broad set of approaches to safety evaluation. In this report, we summarise and share elements of our evolvin...
213
[ "Laura Weidinger", "Joslyn Barnhart", "Jenny Brennan", "Christina Butterfield", "Susie Young", "Will Hawkins", "Lisa Anne Hendricks", "Ramona Comanescu", "Oscar Chang", "Mikel Rodriguez", "Jennifer Beroshi", "Dawn Bloxwich", "Lev Proleev", "Jilin Chen", "Sebastian Farquhar", "Lewis Ho"...
19
Laura Weidinger
[]
cs.AI
[ "cs.AI", "cs.LG" ]
true
2024-04-22T10:26:49
2024-04-22T00:00:00
2024-04-22T10:26:49
null
null
null
10 pages excluding bibliography
https://arxiv.org/abs/2404.14068v1
https://arxiv.org/pdf/2404.14068v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2404.12754v1
2404.12754
1
Adaptive Regularization of Representation Rank as an Implicit Constraint of Bellman Equation
Representation rank is an important concept for understanding the role of Neural Networks (NNs) in Deep Reinforcement learning (DRL), which measures the expressive capacity of value networks. Existing studies focus on unboundedly maximizing this rank; nevertheless, that approach would introduce overly complex models in...
193
[ "Qiang He", "Tianyi Zhou", "Meng Fang", "Setareh Maghsudi" ]
4
Qiang He
[]
cs.LG
[ "cs.LG", "cs.AI" ]
true
2024-04-19T10:00:34
2024-04-19T00:00:00
2024-04-19T10:00:34
null
null
null
Accepted to CVPR23; Code: https://github.com/sweetice/BEER-ICLR2024
https://arxiv.org/abs/2404.12754v1
https://arxiv.org/pdf/2404.12754v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9
2404.11834v1
2404.11834
1
Actor-Critic Reinforcement Learning with Phased Actor
Policy gradient methods in actor-critic reinforcement learning (RL) have become perhaps the most promising approaches to solving continuous optimal control problems. However, the trial-and-error nature of RL and the inherent randomness associated with solution approximations cause variations in the learned optimal valu...
218
[ "Ruofan Wu", "Junmin Zhong", "Jennie Si" ]
3
Ruofan Wu
[]
cs.LG
[ "cs.LG" ]
false
2024-04-18T01:27:31
2024-04-18T00:00:00
2024-04-18T01:27:31
null
null
null
null
https://arxiv.org/abs/2404.11834v1
https://arxiv.org/pdf/2404.11834v1
(all:"Google DeepMind" OR all:DeepMind)
2026-08-02T22:38:49
dFKXatlrhbN1bFDIw
XYgKrIazvJEctQUl9