Jacob Andreas
ITT Career Development Professor in Computer Technology; Associate Professor of EECS
Studies the computational foundations of language learning, machine learning from natural-language guidance, and tools for interpreting black-box models.
Jacob Andreas's lab at MIT works in Natural Language Processing. 70 papers since 2023, 39 at top venues, mostly ICLR and ICML. 9 PhD students. 49 alumni. Applications close .
At a glance
- papers, 2023–2026
- 18/ year
- at T1 venues
- 56%
- PhD students
- 9
- MS students
- 0
- alumni listed
- 49
Papers by yearfrom the lab homepage, 2023–2026
- 14 papers2026 · T1 12 · collecting
- 13 papers2025 · T1 8
- 25 papers2024 · T1 13
- 18 papers2023 · T1 6
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, EMNLP, Findings of ACL
Research focuspaper titles that use each topic's words
- language learning1 paper
- computational1 paper
- foundations0 papers
- tools0 papers
- electrical engineering0 papers
Paper list
From the lab homepage.
202614 papers
- CVPR (T1)ARC is a vision problem! (opens in a new tab)
- ICLR (T1)Beyond binary rewards: training LMs to reason about their uncertainty. (opens in a new tab)
- NeurIPS (T1)Embedded language flows. (opens in a new tab)
- NeurIPS (T1)Introspective coupling: Self-explanation training tracks behavioral change despite fixed supervision. (opens in a new tab)
- ICLR (T1)Language and experience: A computational model of social learning in complex tasks. (opens in a new tab)
- ICLR (T1)Log-augmented generation: Scaling test-time reasoning with reusable computation. (opens in a new tab)
- AIED .Modeling student learning with 3.8 million program traces. (opens in a new tab)
- ICLR (T1)OpenEstimate: Evaluating LLMs on reasoning under uncertainty with real-world data. (opens in a new tab)
- TMLR (T1)Policy learning with a language bottleneck. (opens in a new tab)
- ICML (T1)Position: It’s time to optimize for self-consistency. (opens in a new tab)
- ICML (T1)Reaching beyond the mode: RL for distributional reasoning in language models. (opens in a new tab)
- ICLR (T1)Sequential–parallel duality in prefix scannable models. (opens in a new tab)
202513 papers
- ICML (T1)(How) do language models track state? (opens in a new tab)
- ICML (T1)A hitchhiker’s guide to scaling law estimation. (opens in a new tab)
- NAACL (T1)A probabilistic framework for LLM hallucination detection via belief tree propagation. (opens in a new tab)
- NeurIPS (T1)A spectral understanding of LoRA fine-tuning. (opens in a new tab)
- COLM .Breakpoint: Scalable evaluation of system-level reasoning in LLM agents. (opens in a new tab)
- TACL (T1)Elements of World Knowledge (EWoK): A cognition-inspired framework for evaluating basic world knowledge in language models. (opens in a new tab)
- ICLR (T1)Eliciting human preferences with language models. (opens in a new tab)
- Findings of NAACLLanguage modeling with editable external knowledge. (opens in a new tab)
- ICLR (T1)Learning how hard to think: Input-adaptive allocation of LM computation. (opens in a new tab)
- NeurIPS Oral.Learning linear attention in polynomial time. (opens in a new tab)
- CogSci .Modeling open-world cognition as on-demand synthesis of probabilistic models. (opens in a new tab)
- COLM .Self-steering language models. (opens in a new tab)
202425 papers
- ICML (T1)A multimodal automated interpretability agent. (opens in a new tab)
- CoRL (T1)Adaptive language-guided abstraction from contrastive explanations. (opens in a new tab)
- NeurIPS (T1)Algorithmic capabilities of random transformers. (opens in a new tab)
- COLM .An incomplete loop: Deductive, inductive and abductive learning in large language models. (opens in a new tab)
- Nature Communications (T1)Contextual and combinatorial structure in sperm whale vocalizations. (opens in a new tab)
- ICML Oral.Decomposing uncertainty for large language models through input clarification ensembling. (opens in a new tab)
- Findings of ACLDeductive closure training of language models for coherence, accuracy and updatability. (opens in a new tab)
- EMNLP (T1)Generative AI in the era of “alternative facts”. (opens in a new tab)
- ICML (T1)In-context language learning: Architectures and algorithms. (opens in a new tab)
- COLM .Inspecting and editing knowledge representations in language models. (opens in a new tab)
- ICLR WorkshopLaMPP: Language models as probabilistic priors for perception and action. (opens in a new tab)
- ICLR (T1)Learning adaptive planning representations with natural language guidance. (opens in a new tab)
202318 papers
- Press: MIT News. projectA function interpretation benchmark for evaluating interpretability methods. (opens in a new tab)
- CoNLLAlignment via mutual information. (opens in a new tab)
- Findings of EMNLPAutoReply: Detecting nonsense in dialogue introspectively with discriminative replies. (opens in a new tab)
- —Characterizing intrinsic compositionality in transformers with tree projections. (opens in a new tab)
- EMNLP (T1)Cognitive dissonance: Why do language model outputs disagree with internal representations of truthfulness? (opens in a new tab)
- —Compositional semantic parsing with large language models. (opens in a new tab)
- ACL Lexical Semantics Area Award.Compositionality as lexical symmetry. (opens in a new tab)
- ACL (T1)Grokking of hierarchical structure in vanilla transformers. (opens in a new tab)
- ICML (T1)Guiding pretraining in reinforcement learning with large language models. (opens in a new tab)
- Findings of ACLLanguage modeling with latent situations. (opens in a new tab)
- Neurobiology of Language .Lexical semantic content, not syntactic structure, is the main contributor to ANN-brain similarity of fMRI responses in the language network. (opens in a new tab)
- ICLR Notable Top-5% Paper.Mastering the game of no-press Diplomacy via human-regularized reinforcement learning and planning. (opens in a new tab)
Where alumni went49 listed
Apply to MITFall 2027
MIT EECS Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- TOEFL/IELTS
- Fee
- USD 90
Confirmed on the program page.
Similar labsnearest by research summary
- CNPeking
Sujian Li
- Dialogue, generation & summarization
- USUIUC
Dilek Hakkani-Tur
- Dialogue, generation & summarization
- USMichigan
Emily Mower Provost
- Speech recognition & spoken language
- USStanford
Dan Jurafsky
- Language & society
- GBEdinburgh
Zee Talat
- Recognition, detection & segmentation
- USUMD
Jordan Lee Boyd-Graber
- Knowledge, QA & retrieval
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
MITcloses