Matthew So

Amogh Inamdar, Matthew So, Vici Milenia et al.

Continual learning methods aim to maximize the stability and plasticity of machine learning models that are trained on a sequence of tasks. The standard measure of stability (i.e., forgetting) is the 0-shot performance of a model on previously learned tasks, and plasticity, the performance on the most recently learned task. However, 0-shot evaluation does not fully measure a model or method's ability to retain learned information or adapt quickly to new information, as it requires perfect recall across multiple tasks. In this paper, we propose few-shot evaluation as a more comprehensive assessment of the stability and plasticity of a continual learning system. We conduct a fine-grained assessment on task sequences for continual image classification and find that this paradigm produces novel insights into the performance of popular continual learning strategies. Through few-shot evaluation with a novel metric -- per-shot plasticity -- we show that adding `foresight' to continual learning methods via the meta-learning of a short sequence of future tasks induces learning-to-learn behavior over the task sequence.

Semyon Lomasov, Judah Goldfeder, Mehmet Hamza Erol et al.

Do AI systems truly understand human concepts or merely mimic surface patterns? We investigate this through chess, where human creativity meets precise strategic concepts. Analyzing a 270M-parameter transformer that achieves grandmaster-level play, we uncover a striking paradox: while early layers encode human concepts like center control and knight outposts with up to 85\% accuracy, deeper layers, despite driving superior performance, drift toward alien representations, dropping to 50-65\% accuracy. To test conceptual robustness beyond memorization, we introduce the first Chess960 dataset: 240 expert-annotated positions across 6 strategic concepts. When opening theory is eliminated through randomized starting positions, concept recognition drops 10-20\% across all methods, revealing the model's reliance on memorized patterns rather than abstract understanding. Our layer-wise analysis exposes a fundamental tension in current architectures: the representations that win games diverge from those that align with human thinking. These findings suggest that as AI systems optimize for performance, they develop increasingly alien intelligence, a critical challenge for creative AI applications requiring genuine human-AI collaboration. Dataset and code are available at: https://github.com/slomasov/ChessConceptsLLM.

Xiao Yu, Ruize Xu, Chengyuan Xue et al.

A reliable resume-job matching system helps a company find suitable candidates from a pool of resumes and helps a job seeker find relevant jobs from a list of job posts. While recent advances in embedding-based methods such as ConFit and ConFit v2 can efficiently retrieve candidates at scale, the lack of controllability and explainability limits their real-world adaptations. LLM-based re-rankers can address these limitations through reasoning, but existing training recipes are developed on short-document benchmarks and do not account for noise in real-world recruiting data. In this work, we first conduct a systematic analysis over the LLM re-ranker training pipeline for person-job fit, covering inference algorithm design, RL algorithm selection, data processing, and SFT distillation. We find that using multi-pass re-ranking, training with listwise RL objectives, removing noisy samples, and distilling from a stronger LLM before RL significantly improves re-ranking performance. We then aggregate these findings to train ConFit v3 with Qwen3-8B and Qwen3-32B on real-world person-job fit datasets, and find significant improvements over existing best person-job fit systems as well as strong LLMs such as GPT-5 and Claude Opus-4.5. We hope our findings provide useful insights for future research on adapting LLM-based re-rankers to person-job fit systems.

Matthew So, Judah Goldfeder, Mark Lis et al.

There has been a historic assumption that the biometrics of an individual are statistically uncorrelated. We test this assumption by training Bi-Encoder networks on three verification tasks, including fingerprint-to-fingerprint matching, iris-to-iris matching, and cross-modal fingerprint-to-iris matching using 274 subjects with $\sim$100k fingerprints and 7k iris images. We trained ResNet-50 and Vision Transformer backbones in Bi-Encoder architectures such that the contrastive loss between images sampled from the same individual is minimized. The iris ResNet architecture reaches 91 ROC AUC score for iris-to-iris matching, providing clear evidence that the left and right irises of an individual are correlated. Fingerprint models reproduce the positive intra-subject suggested by prior work in this space. This is the first work attempting to use Vision Transformers for this matching. Cross-modal matching rises only slightly above chance, which suggests that more data and a more sophisticated pipeline is needed to obtain compelling results. These findings continue challenge independence assumptions of biometrics and we plan to extend this work to other biometrics in the future. Code available: https://github.com/MatthewSo/bio_fingerprints_iris.

Judah Goldfeder, Matthew So, Hod Lipson

Auxiliary Learning (AL) is a form of multi-task learning in which a model trains on auxiliary tasks to boost performance on a primary objective. While AL has improved generalization across domains such as navigation, image classification, and NLP, it often depends on human-labeled auxiliary tasks that are costly to design and require domain expertise. Meta-learning approaches mitigate this by learning to generate auxiliary tasks, but typically rely on gradient based bi-level optimization, adding substantial computational and implementation overhead. We propose RL-AUX, a reinforcement-learning (RL) framework that dynamically creates auxiliary tasks by assigning auxiliary labels to each training example, rewarding the agent whenever its selections improve the performance on the primary task. We also explore learning per-example weights for the auxiliary loss. On CIFAR-100 grouped into 20 superclasses, our RL method outperforms human-labeled auxiliary tasks and matches the performance of a prominent bi-level optimization baseline. We present similarly strong results on other classification datasets. These results suggest RL is a viable path to generating effective auxiliary tasks.