Pith. sign in

REVIEW 4 cited by

Combining Modular Skills in Multitask 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 2202.13914 v2 pith:NRAH2UE6 submitted 2022-02-28 cs.LG cs.CL

Combining Modular Skills in Multitask Learning

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

A modular design encourages neural models to disentangle and recombine different facets of knowledge to generalise more systematically to new tasks. In this work, we assume that each task is associated with a subset of latent discrete skills from a (potentially small) inventory. In turn, skills correspond to parameter-efficient (sparse / low-rank) model parameterisations. By jointly learning these and a task-skill allocation matrix, the network for each task is instantiated as the average of the parameters of active skills. To favour non-trivial soft partitions of skills across tasks, we experiment with a series of inductive biases, such as an Indian Buffet Process prior and a two-speed learning rate. We evaluate our latent-skill model on two main settings: 1) multitask reinforcement learning for grounded instruction following on 8 levels of the BabyAI platform; and 2) few-shot adaptation of pre-trained text-to-text generative models on CrossFit, a benchmark comprising 160 NLP tasks. We find that the modular design of a network significantly increases sample efficiency in reinforcement learning and few-shot generalisation in supervised learning, compared to baselines with fully shared, task-specific, or conditionally generated parameters where knowledge is entangled across tasks. In addition, we show how discrete skills help interpretability, as they yield an explicit hierarchy of tasks.

discussion (0)

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

Forward citations

Cited by 4 Pith papers

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

  1. ARIADNE: Agnostic Routing for Inference-time Adapter DyNamic sElection

    cs.AI 2026-06 unverdicted novelty 6.0

    ARIADNE routes queries to the best adapter via embedding-space centroid proximity, recovering 97.44% of upper-bound performance on 23 NLP tasks and 89.7% selection accuracy on 44 tasks without training or internal access.

  2. HMR-Net: Hierarchical Modular Routing for Cross-Domain Object Detection in Aerial Images

    cs.CV 2026-04 unverdicted novelty 6.0

    HMR-Net introduces hierarchical routing with global dataset-level and local scene-level modularity plus conditional experts to improve cross-domain aerial object detection and enable novel category recognition without...

  3. DoRA: Weight-Decomposed Low-Rank Adaptation

    cs.CL 2024-02 accept novelty 6.0

    DoRA improves LoRA by decomposing weights into magnitude and direction and updating only direction with low-rank matrices, closing much of the gap to full fine-tuning.

  4. Modular Foundation Models for Time-Series Perception in Digital Twins

    cs.LG 2026-07 conditional novelty 5.0

    A gated bank of frozen self-supervised time-series encoders, aligned and aggregated by a Transformer, supports competitive multi-task perception for digital twins and hydro-generator virtual sensing.