Pith. sign in

REVIEW 5 cited by

Conflict-Averse Gradient Descent for Multi-task Learning

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2110.14048 v2 pith:EZHUFD3Z submitted 2021-10-26 cs.LG cs.AI

Conflict-Averse Gradient Descent for Multi-task Learning

classification cs.LG cs.AI
keywords learninggradientmulti-tasktasksaveragedescenttaskcagrad
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The goal of multi-task learning is to enable more efficient learning than single task learning by sharing model structures for a diverse set of tasks. A standard multi-task learning objective is to minimize the average loss across all tasks. While straightforward, using this objective often results in much worse final performance for each task than learning them independently. A major challenge in optimizing a multi-task model is the conflicting gradients, where gradients of different task objectives are not well aligned so that following the average gradient direction can be detrimental to specific tasks' performance. Previous work has proposed several heuristics to manipulate the task gradients for mitigating this problem. But most of them lack convergence guarantee and/or could converge to any Pareto-stationary point. In this paper, we introduce Conflict-Averse Gradient descent (CAGrad) which minimizes the average loss function, while leveraging the worst local improvement of individual tasks to regularize the algorithm trajectory. CAGrad balances the objectives automatically and still provably converges to a minimum over the average loss. It includes the regular gradient descent (GD) and the multiple gradient descent algorithm (MGDA) in the multi-objective optimization (MOO) literature as special cases. On a series of challenging multi-task supervised learning and reinforcement learning tasks, CAGrad achieves improved performance over prior state-of-the-art multi-objective gradient manipulation methods.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Biological Reasoning-Informed Regression for Interpretable Regulatory DNA Activity Prediction

    q-bio.GN 2026-06 unverdicted novelty 6.0

    R3LM trains LLMs via two-stage reasoning-then-regression on a new dataset CRE-ReasonBench with mechanistic traces, achieving SOTA enhancer activity prediction across three cell types with interpretable outputs.

  2. 21cmEMUv3: a hybrid diffusion-LSTM emulator of 21cmFAST summary observables

    astro-ph.CO 2026-05 unverdicted novelty 6.0

    21cmEMUv3 emulates the cylindrical 21cm power spectrum via score-based diffusion and six other 21cmFAST observables via LSTM networks at sub-percent accuracy, then uses the emulator to infer a lower limit on soft-band...

  3. CoAction: Cross-task Correlation-aware Pareto Set Learning

    cs.LG 2026-05 unverdicted novelty 6.0

    CoAction applies a transformer encoder with per-task embeddings to jointly solve multiple multi-objective optimization problems by capturing cross-task correlations.

  4. CoAction: Cross-task Correlation-aware Pareto Set Learning

    cs.LG 2026-05 unverdicted novelty 6.0

    CoAction introduces a task-aware transformer that simultaneously learns Pareto optimal solutions across multiple tasks by capturing inter-task correlations via self-attention and task embeddings.

  5. Multi-Task Learning for Heterogeneous Prediction from Video Game State with Transfer Learning

    cs.LG 2026-07 conditional novelty 4.0

    On a large World of Tanks dataset, a shared multi-task model with equal weighting or PCGrad outperforms single-task models on average, and task/map pre-training helps most in low-data regimes.