Real-Time Execution with Autoregressive Policies
The Conference on Robot Learning
CoRL
Avi Caciularu
Senior Research Scientist @ Google Research
About Me
I am a Senior Research Scientist at Google Research, where I focus on making AI act rather than just speak — building agents that reason through code and collaborate across extended multi-turn workflows.
My academic journey includes completing a Ph.D. in Computer Science at the BIU NLP Lab, where I had the privilege of working with Prof. Ido Dagan, Prof. Jacob Goldberger, and Dr. Arman Cohan (from Yale). My previous education includes an M.Sc in Electrical Engineering under the supervision of Prof. David Burshtein, and a B.Sc in Computer Science and Electrical Engineering, both from Tel-Aviv University.
My research focuses on advancing reasoning and interactive agents by enhancing models' ability to plan, use tools, and execute decisions in software and interactive environments. In parallel, I am interested in studying model uncertainty, reliability, and interpretability.
I am always open to discussions and potential collaborations on these topics. If you have insights, questions, or are interested in exploring these areas together, please feel free to reach out.
Résumé
Education
Experience
Publications
2026
Latent Reasoning with Supervised Thinking States
The Conference on Language Modeling, Workshop on Efficient Reasoning
COLM
Fine-Tuning Was Not Broken but Diffuse: Precise Knowledge Editing via Projected Error Signals
The Conference on Language Modeling
COLM
Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs
The Conference on Language Modeling
COLM
ConvApparel: A Benchmark Dataset and Validation Framework for User Simulators in Conversational Recommenders
The Conference of the European Chapter of the Association for Computational Linguistics
EACL
2025
DoubleDipper: Recycling Contexts for Efficient and Attributed In-Context Learning
The International Joint Conference on Natural Language Processing & Asia-Pacific Chapter of the ACL, Findings
AACL Findings
MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs
The Conference on Empirical Methods in Natural Language Processing
EMNLP
ReliableEval: A Recipe for Stochastic LLM Evaluation via Method of Moments
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
Technical Report
MDCure: A Scalable Pipeline for Multi-Document Instruction-Following
The Annual Meeting of the Association for Computational Linguistics
ACL
Identifying User Goals From UI Trajectories
The Web Conference (WWW), Workshop on Personal Intelligence with Generative AI
WWW
Best Paper Award
2024
TACT: Advancing Complex Aggregative Reasoning with Information Extraction Tools
Annual Conference on Neural Information Processing Systems (Datasets and Benchmarks)
NeurIPS
Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach
Technical Report
Unpacking Tokenization: Evaluating Text Compression and its Correlation with Model Performance
The Annual Meeting of the Association for Computational Linguistics, Findings
ACL Findings
Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Models
The International Conference on Machine Learning
ICML
2023
Stop Uploading Test Data in Plain Text: Practical Strategies for Mitigating Data Contamination by Evaluation Benchmarks
The Conference on Empirical Methods in Natural Language Processing
EMNLP
The Curious Case of Hallucinatory Unanswerablity: Finding Truths in the Hidden States of Over-Confident Large Language Models
The Conference on Empirical Methods in Natural Language Processing
EMNLP
Optimizing Retrieval-augmented Reader Models via Token Elimination
The Conference on Empirical Methods in Natural Language Processing
EMNLP
A Comprehensive Evaluation of Tool-Assisted Generation Strategies
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
Don't Add, don't Miss: Effective Content Preserving Generation from Pre-Selected Text Spans
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
Peek Across: Improving Multi-Document Modeling via Cross-Document Question-Answering
The Annual Meeting of the Association for Computational Linguistics
ACL
Revisiting Sentence Union Generation as a Testbed for Text Consolidation
The Annual Meeting of the Association for Computational Linguistics, Findings
ACL Findings
An Entangled Mixture of Variational Autoencoders Approach to Deep Clustering
Neurocomputing
Journal
Explaining the decisions of power quality disturbance classifiers using latent space features
The International Journal of Electrical Power and Energy Systems
Journal
2022
Cross-document Event Coreference Search: Task, Dataset and Modeling
The Conference on Empirical Methods in Natural Language Processing
EMNLP
QASem Parsing: Text-to-text Modeling of QA-based Semantics
The Conference on Empirical Methods in Natural Language Processing
EMNLP
Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space
The Conference on Empirical Methods in Natural Language Processing
EMNLP
LM-Debugger: An Interactive Tool for Inspection and Intervention in Transformer-Based Language Models
The Conference on Empirical Methods in Natural Language Processing, System Demonstrations
EMNLP Demo
Long Context Question Answering via Supervised Contrastive Learning
The Annual Conference of the North American Chapter of the ACL
NAACL
A Proposition-Level Clustering Approach for Multi-Document Summarization
The Annual Conference of the North American Chapter of the ACL
NAACL
Interpreting BERT-based Text Similarity via Activation and Saliency Maps
The World Wide Web Conference
WWW
MetricBERT: Text Representation Learning via Self-Supervised Triplet Training
The International Conference on Acoustics, Speech, and Signal Processing
ICASSP
2021
CDLM: Cross-document Language Modeling
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
iFacetSum: Coreference-based Interactive Faceted Summarization for Multi-Document Exploration
The Conference on Empirical Methods in Natural Language Processing, System Demonstrations
EMNLP Demo
Cold Item Integration in Deep Hybrid Recommenders via Tunable Stochastic Gates
The International Conference on Data Mining
ICDM
Representation Learning via Variational Bayesian Networks
The International Conference on Information and Knowledge Management
CIKM
Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps
The International Conference on Information and Knowledge Management
CIKM
Anchor-based Collaborative Filtering
The International Conference on Information and Knowledge Management
CIKM
GAM: Explainable Visual Similarity and Classification via Gradient Activation Maps
The International Conference on Information and Knowledge Management
CIKM
On the Evolution of Word Order
Recent Advances in Natural Language Processing (RANLP), Student Research Workshop
RANLP
Denoising Word Embeddings by Averaging in a Shared Space
The Joint Conference on Lexical and Computational Semantics
*SEM
Self-Supervised Document Similarity Ranking via Contextualized Language Models and Hierarchical Inference
The Annual Meeting of the Association for Computational Linguistics, Findings
ACL Findings
Cold Start Revisited: A Deep Hybrid Recommender with Cold-Warm Item Harmonization
The International Conference on Acoustics, Speech, and Signal Processing
ICASSP
perm2vec: Attentive Graph Permutation Selection for Decoding of Error Correction Codes
IEEE Transactions on Cognitive Communications and Networking, Special Issue
IEEE TCCN
2020
Within-Between Lexical Relation Classification
The Conference on Empirical Methods in Natural Language Processing
EMNLP
Paraphrasing vs Coreferring: Two Sides of the Same Coin
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
RecoBERT: A Catalog Language Model for Text-Based Recommendations
The Conference on Empirical Methods in Natural Language Processing, Findings
EMNLP Findings
Cold Item Recommendations via Hierarchical Item2vec
The International Conference on Data Mining
ICDM
Explainable Recommendations via Attentive Multi-Persona Collaborative Filtering
The ACM Conference on Recommender Systems
RecSys
Attentive Item2vec: Neural Attentive User Representations
The International Conference on Acoustics, Speech, and Signal Processing
ICASSP
Bayesian Hierarchical Words Representation Learning
The Annual Meeting of the Association for Computational Linguistics
ACL
Unsupervised Linear and Nonlinear Channel Equalization and Decoding using Variational Autoencoders
IEEE Transactions on Cognitive Communications and Networking
IEEE TCCN
Scalable Attentive Sentence-Pair Modeling via Distilled Sentence Embedding
The AAAI Conference on Artificial Intelligence
AAAI
2018
ARPM: Additive, Retentive Penalty Method for Multidimensional NILM Algorithms
The International Conference on the Science of Electrical Engineering
ICSEE
Blind Channel Equalization Using Variational Autoencoders
The International Conference on Communications
ICC
Inducing Regular Grammars Using Recurrent Neural Networks
The International Joint Conferences on Artificial Intelligence (IJCAI), Workshop on Learning and Reasoning
IJCAI