Papers by Tommi Jaakkola
Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control (D19-1)
Copied to clipboard
| Challenge: | Selective rationalization is a common mechanism to ensure that predictive models reveal how they use any available features. |
| Approach: | They propose a co-operative method which uses introspection to explicitly predict and incorporate the outcome into the selection process. |
| Outcome: | The proposed model maintains high predictive accuracy and leads to comprehensive rationales. |
Consistent Accelerated Inference via Confident Adaptive Transformers (2021.emnlp-main)
Copied to clipboard
| Challenge: | Amortized or approximate computational methods increase efficiency, but can result in unpredictable performance costs. |
| Approach: | They propose a method that increases computational efficiency while guaranteeing a specifiable degree of consistency with the original model with high confidence. |
| Outcome: | The proposed method improves on four classification and regression tasks and can be used to predict the performance of the proposed model. |
Blank Language Models (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches focus on adapting left-to-right language models for text infilling. |
| Approach: | They propose a model that generates sequences by dynamically creating and filling in blanks. |
| Outcome: | Experiments on style transfer and damaged ancient text restoration show that the proposed model outperforms baseline models in terms of accuracy and fluency. |
Gromov-Wasserstein Alignment of Word Embedding Spaces (D18-1)
Copied to clipboard
| Challenge: | Current unsupervised methods for learning cross-lingual correspondences involve multiple steps, including heuristic post-hoc refinement strategies. |
| Approach: | They propose to cast the correspondence problem directly as an optimal transport problem, building on the idea that word embeddings arise from metric recovery algorithms. |
| Outcome: | The proposed method can be estimated efficiently, requires little or no tuning, and performs comparable with the state-of-the-art in various unsupervised word translation tasks. |
Revisiting Who’s Harry Potter: Towards Targeted Unlearning from a Causal Intervention Perspective (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing and new datasets show that our approach achieves competitive performance in all of the criteria. |
| Approach: | They propose a new task of LLM targeted unlearning where unlearning targets only the information about the unlearning target, rather than everything in the unlearned documents. |
| Outcome: | The proposed method achieves competitive performance on existing and new datasets without optimizing for the aforementioned criteria. |
Thought calibration: Efficient and confident test-time scaling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for teaching language models to be economical with their token budgets have failed to achieve the desired results. |
| Approach: | They propose to calibrate a language model's growing body of thoughts to determine when new reasoning plateaus. |
| Outcome: | The proposed framework preserves model performance with up to 60% reduction in thinking tokens on in-distribution data, and up to 20% in out-of-difference data. |