SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.05932 · ICLR · 2025

Skill Expansion and Composition in Parameter Space

Haoyi Niu, Xianyuan Zhan, Jianxiong Li, Yinan Zheng, Xin Xu, Tenglong Liu, Yixing Lan

arXiv · PDF · Open in the Atlas

Code that ran

10 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 2 violated it and 2 have no contract verdict; 2 failures of our instrument; 2 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2502.05932.json · badge: /api/ran/2502.05932.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
ltlhuuu/PSEC canonical (classifier label)
classifier basis not recorded on the link
10 ran · 2 no recorded run
FunctionStatusWhere it lives
compute_q Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ltlhuuu/PSEC/jaxrl5/agents/psec/psec_finetune.py
pointer only (licence: MIT); the stored copy get_code serves, from ZhengYinan-AIR/FISOR, is pointer only (licence: NONE) · get_code("efbad980c5e8e094")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ZhengYinan-AIR/FISOR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

cosine_beta_schedule Ran
ran, honoured the contract we drafted
ltlhuuu/PSEC/jaxrl5/networks/diffusion.py
pointer only (licence: MIT); the stored copy get_code serves, from Alescontrela/score_matching_rl, is pointer only (licence: NONE) · get_code("87d1bc442d93916d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from Alescontrela/score_matching_rl, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

evaluate1 Ran
ran on a synthesized input
ltlhuuu/PSEC/jaxrl5/evaluation_dsrl.py
code served (permissive licence) · get_code("74b0ae56402911be")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
import gym
from gym import spaces
from unittest.mock import Mock

torch.manual_seed(0)
np.random.seed(0)

# Create a minimal gym environment
class MinimalEnv(gym.Env):
    def __init__(self):
        self.observation_space = spaces.Box(low=-1, high=1, shape=(4,), dtype=np.float32)
        self.action_space = spaces.Box(low=-1, high=1, shape=(2,), dtype=np.float32)
        self.step_count = 0
    
    def reset(self):
        self.step_count = 0
        obs = self.observation_space.sample()
        info = {}
        return obs, info
    
    def step(self, action):
        self.step_count += 1
        obs = self.observation_space.sample()
        reward = float(np.random.randn())
        terminated = self.step_count >= 3
        truncated = False
        info = {'cost': float(np.random.rand())}
        return obs, reward, terminated, truncated, info
    
    def render(self):
        pass

# Create a minimal agent with eval_actions method
class MinimalAgent:
    def eval_actions(self, obs, train_lora=True, eval_maxq_weight=0.07, eval_minqc_weight=0.07):
        action = np.array([0.1, 0.2], dtype=np.float32)
        q1_eps_pred_lora_dis = np.array([0.5, 0.6], dtype=np.float32)
        q2_eps_pred_lora_dis = np.array([0.4, 0.7], dtype=np.float32)
        eps_pred_dis = np.array([0.3, 0.8], dtype=np.float32)
        return action, self, q1_eps_pred_lora_dis, q2_eps_pred_lora_dis, eps_pred_dis

agent = MinimalAgent()
env = MinimalEnv()

ARGS = {
    'agent': agent,
    'env': env,
    'num_episodes': 2,
    'save_video': False,
    'render': False,
    'train_lora': True,
    'eval_maxq_weight': 0.07,
    'eval_minqc_weight': 0.07,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
gym, numpy, torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
evaluate2 Ran
ran on a synthesized input
ltlhuuu/PSEC/jaxrl5/evaluation_dsrl.py
code served (permissive licence) · get_code("c7143d30af7311f2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
import gym
from gym import spaces
from unittest.mock import Mock

torch.manual_seed(0)

# Create a minimal gym.Env
class MinimalEnv(gym.Env):
    def __init__(self):
        self.observation_space = spaces.Box(low=-1, high=1, shape=(4,), dtype=np.float32)
        self.action_space = spaces.Box(low=-1, high=1, shape=(3,), dtype=np.float32)
        self.step_count = 0
    
    def reset(self):
        self.step_count = 0
        obs = self.observation_space.sample()
        info = {}
        return obs, info
    
    def step(self, action):
        self.step_count += 1
        obs = self.observation_space.sample()
        reward = float(np.random.randn())
        terminated = self.step_count >= 5
        truncated = False
        info = {'cost': float(np.random.rand())}
        return obs, reward, terminated, truncated, info
    
    def render(self):
        pass

# Create a minimal agent with eval_actions method
class MinimalAgent:
    def eval_actions(self, obs, train_lora=True, eval_maxq_weight=0.07, eval_minqc_weight=0.07):
        action = np.array([0.1, 0.2, 0.3], dtype=np.float32)
        q1_eps_pred_lora_dis = np.array([1.0, 2.0])
        q2_eps_pred_lora_dis = np.array([1.5, 2.5])
        eps_pred_dis = np.array([0.5, 0.6])
        return action, self, q1_eps_pred_lora_dis, q2_eps_pred_lora_dis, eps_pred_dis

agent = MinimalAgent()
env = MinimalEnv()

ARGS = {
    'agent': agent,
    'env': env,
    'num_episodes': 2,
    'save_video': False,
    'render': False,
    'train_lora': True,
    'eval_maxq_weight': 0.07,
    'eval_minqc_weight': 0.07
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
gym, numpy, torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
expectile_loss Ran
ran, violated the contract we drafted
ltlhuuu/PSEC/jaxrl5/agents/psec/psec_finetune.py
pointer only (licence: MIT); the stored copy get_code serves, from ZhengYinan-AIR/FISOR, is pointer only (licence: NONE) · get_code("ad221b0896e1bcd0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ZhengYinan-AIR/FISOR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_weight_decay_mask Ran
ran; our contract draft was wrong, not the code
ltlhuuu/PSEC/jaxrl5/networks/mlp.py
pointer only (licence: MIT); the stored copy get_code serves, from zhengyinan-air/fisor, is pointer only (licence: NONE) · get_code("1049990078a4b0f2")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from zhengyinan-air/fisor, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

linear_beta_schedule Ran
ran, honoured the contract we drafted
ltlhuuu/PSEC/jaxrl5/networks/diffusion.py
pointer only (licence: MIT); the stored copy get_code serves, from diffusionyes/MaxEntDP, is pointer only (licence: NONE) · get_code("0cd44ae2231fcb3b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from diffusionyes/MaxEntDP, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

mish Ran
ran, honoured the contract we drafted
ltlhuuu/PSEC/jaxrl5/agents/psec/psec_pretrain.py
pointer only (licence: MIT); the stored copy get_code serves, from diffusionyes/maxentdp, is pointer only (licence: NONE) · get_code("7fc1d127bf81537f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from diffusionyes/maxentdp, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

safe_expectile_loss Ran
ran, violated the contract we drafted
ltlhuuu/PSEC/jaxrl5/agents/psec/psec_finetune.py
pointer only (licence: MIT); the stored copy get_code serves, from ZhengYinan-AIR/FISOR, is pointer only (licence: NONE) · get_code("7e821a429b5ff00e")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from ZhengYinan-AIR/FISOR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

vp_beta_schedule Ran
ran, honoured the contract we drafted
ltlhuuu/PSEC/jaxrl5/networks/diffusion.py
pointer only (licence: MIT); the stored copy get_code serves, from Alescontrela/score_matching_rl, is pointer only (licence: NONE) · get_code("539aaf559a17a19a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from Alescontrela/score_matching_rl, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

evaluate Not yet run
unverified (harvested, no recorded run)
ltlhuuu/PSEC/jaxrl5/evaluation_dsrl.py
code served (permissive licence) · get_code("27b1370715ab6f72")
How it ran

no invocation recorded.

subsample_ensemble Not yet run
unverified (harvested, no recorded run)
ltlhuuu/PSEC/jaxrl5/networks/ensemble.py
pointer only (licence: MIT); the stored copy get_code serves, from diffusionyes/MaxEntDP, is pointer only (licence: NONE) · get_code("c021b9fd8bb7677d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from diffusionyes/MaxEntDP, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 9 are also present in other papers whose code we hold (9 with at least one paper dated earlier, 4 the same month, 5 later) and 3 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
mish 7 other papers
4 dated earlier, 1 the same month, 2 later · 1 only as an assembly brick
2312.11752 (2023-12)
Learning a Diffusion Model Policy from Rewards via Q-Score Matching
2402.17135 (2024-02)
Unsupervised Zero-Shot Reinforcement Learning via Functional Reward Encodings
2405.20555 (2024-05)
Diffusion Actor-Critic: Formulating Constrained Policy Iteration as Diffusion Noise Regression for Offline Reinforcement Learning
and 4 more not named here
cosine_beta_schedule 5 other papers
3 dated earlier, 1 the same month, 1 later
2308.12952 (2023-08)
BridgeData V2: A Dataset for Robot Learning at Scale
2312.11752 (2023-12)
Learning a Diffusion Model Policy from Rewards via Q-Score Matching
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
and 2 more not named here
vp_beta_schedule 5 other papers
3 dated earlier, 1 the same month, 1 later
2308.12952 (2023-08)
BridgeData V2: A Dataset for Robot Learning at Scale
2312.11752 (2023-12)
Learning a Diffusion Model Policy from Rewards via Q-Score Matching
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
and 2 more not named here
linear_beta_schedule 4 other papers
2 dated earlier, 1 the same month, 1 later
2308.12952 (2023-08)
BridgeData V2: A Dataset for Robot Learning at Scale
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
2502.05450 (2025-02)
ConRFT: A Reinforced Fine-tuning Method for VLA Models via Consistency Policy
and 1 more not named here
expectile_loss 3 other papers
3 dated earlier
2304.10573 (2023-04)
IDQL: Implicit Q-Learning as an Actor-Critic Method with Diffusion Policies
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
2402.17135 (2024-02)
Unsupervised Zero-Shot Reinforcement Learning via Functional Reward Encodings
subsample_ensemble 3 other papers
2 dated earlier, 1 later
2311.05067 (2023-11)
Accelerating Exploration with Unlabeled Prior Data
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
2605.21282 (2026-05)
Stochastic MeanFlow Policies: One-Step Generative Control with Entropic Mirror Descent
compute_q 1 other paper
1 dated earlier
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
get_weight_decay_mask 1 other paper
1 dated earlier
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
safe_expectile_loss 1 other paper
1 dated earlier
2401.10700 (2024-01)
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Humans excel at reusing prior knowledge to address new challenges and developing skills while solving problems. This paradigm becomes increasingly popular in the development of autonomous agents, as it develops systems that can self-evolve in response to new challenges like human beings. However, previous methods suffer from limited training efficiency when expanding new skills and fail to fully leverage prior knowledge to facilitate new task learning. In this paper, we propose Parametric Skill Expansion and Composition (PSEC), a new framework designed to iteratively evolve the agents' capabilities and efficiently address new challenges by maintaining a manageable skill library. This library can progressively integrate skill primitives as plug-and-play Low-Rank Adaptation (LoRA) modules in parameter-efficient finetuning, facilitating efficient and flexible skill expansion. This structure also enables the direct skill compositions in parameter space by merging LoRA modules that encode different skills, leveraging shared information across skills to effectively program new skills. Based on this, we propose a context-aware module to dynamically activate different skills to collaboratively handle new tasks. Empowering diverse applications including multi-objective composition, dynamics shift, and continual policy shift, the results on D4RL, DSRL benchmarks, and the DeepMind Control Suite show that PSEC exhibits superior capacity to leverage prior knowledge to efficiently tackle new challenges, as well as expand its skill libraries to evolve the capabilities. Project website: https://ltlhuuu.github.io/PSEC/.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2502.05932")
get_code_for_paper("2502.05932")
have("2502.05932")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2502.05932.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2502.05932.svg)](https://syntology.ai/paper/2502.05932)

Connect an agent — have() is free.