AI Paper Atlas

Research map / Learning theory and algorithms / Combinatorial and search optimization

HOBIT: Hardness Optimized Batch Sampling for InfoNCE Training

Himanshu Dutta, Lokesh Nagalapatti, Yashoteja Prabhu

ICML 2026 · Learning theory and algorithms › Combinatorial and search optimization · approach: Constrained neural network optimization

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use HOBIT batching instead of Random batchingdual encoder (roberta-base) · MSMARCO PassageMRR@10: 24.4 -> 27.4Improved
use HOBIT batching instead of Random batchingdual encoder (roberta-base) · NQRecall@10: 45.1 -> 52.4Improved
use HOBIT batching instead of NGAMEdual encoder (e5-large-unsupervised) · MSMARCO PassageMRR@10: 28.6 -> 30.1Improved
use HOBIT batching instead of NGAMEdual encoder (e5-large-unsupervised) · NQRecall@100: 90.5 -> 89.9Hurt
use fixed (static) teacher-ordered batches (B3) instead of adaptive batchingdual encoder (roberta-base) · MSMARCO PassageMRR@10: 24.4 -> 23.4Hurt
add HOBIT batching on top of ANCE hard negativesANCE (roberta-base) · MSMARCO PassageMRR@10: 27.33 -> 27.29No clear effect
add HOBIT batching on top of TriSamplerTriSampler (roberta-base) · MSMARCO PassageMRR@10: 26.26 -> 26.65Improved
add HOBIT batching on top of ANCEANCE (roberta-base) · NQRecall@10: 59.27 -> 56.43Hurt
add HOBIT batching on top of ANCEANCE (roberta-base) · NQRecall@100: 84.63 -> 85.41Improved
add HOBIT batching on top of TriSamplerTriSampler (roberta-base) · NQRecall@10: 53.80 -> 57.13Improved
add HOBIT batching on top of ANCEANCE (roberta-base) · TriviaQARecall@100: 86.34 -> 88.34Improved
use cached embeddings (HOBIT-C) instead of per-epoch recomputed embeddings (ablation)HOBIT (roberta-base) · NQRecall@10: 52.4 -> 51.7; 29.4% per-epoch speedupNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy