Zero-Shot Environment Solving with Autoregressive Agents
1. Core Principles of Zero-Shot Learning
Core Principles of Zero-Shot Learning
Zero-shot learning (ZSL) enables models to recognize or classify instances of classes they have never seen during training by leveraging auxiliary information. This capability is particularly valuable in scenarios where labeled data is scarce or when the model must generalize to novel categories dynamically.
Semantic Embedding Spaces
The foundation of ZSL lies in mapping both seen and unseen classes into a shared semantic embedding space. This space typically represents attributes, word vectors, or other high-level descriptors that capture class relationships. Formally, let X denote the input space and A the semantic space. The model learns a mapping function f: X → A such that:
where x_i is an instance from class i and a_i is its corresponding semantic representation. During inference, the model projects an unseen class instance into this space and compares it with the embeddings of unseen classes.
Generalization Through Auxiliary Knowledge
ZSL relies on auxiliary information to bridge the gap between seen and unseen classes. Common approaches include:
- Attribute-based models: Classes are described by human-defined attributes (e.g., "has wings", "is metallic").
- Word vector embeddings: Pretrained language models (e.g., Word2Vec, GloVe) provide semantic relationships between class names.
- Knowledge graphs: Structured ontologies encode hierarchical and relational information about classes.
Inductive vs. Transductive ZSL
Two main paradigms exist for zero-shot learning:
- Inductive ZSL: The model only accesses labeled data from seen classes during training. Generalization to unseen classes relies entirely on their semantic descriptions.
- Transductive ZSL: The model additionally leverages unlabeled instances from unseen classes during training, often improving performance but requiring more sophisticated algorithms.
Autoregressive Agents in ZSL
Autoregressive models, particularly large language models (LLMs), have demonstrated remarkable zero-shot capabilities by conditioning on task descriptions and examples. These agents solve novel problems by:
- Leveraging pretrained knowledge encoded in their parameters.
- Generating solutions step-by-step through sequential prediction.
- Adapting to new tasks through prompt engineering without weight updates.
where x is the input, y is the output sequence, and T is the sequence length. This autoregressive formulation allows the model to tackle novel problems by generating solutions token-by-token based on its pretraining.
Challenges and Limitations
Despite its promise, ZSL faces several key challenges:
- Domain shift: The distribution of features may differ between seen and unseen classes.
- Hubness problem: High-dimensional embedding spaces often suffer from concentration of measure effects.
- Semantic gap: The auxiliary information may not fully capture visual or functional characteristics.
Recent advances address these issues through techniques like generative adversarial networks to synthesize unseen class features, or hybrid models that combine multiple semantic representations.

Transfer Learning and Generalization in Zero-Shot Contexts
Transfer learning enables autoregressive agents to leverage knowledge from previously encountered environments to solve novel tasks without additional training. The core mechanism involves parameter sharing across tasks, where a base model is pretrained on a diverse set of environments, and its learned representations are reused in zero-shot settings. The generalization capability hinges on the model's ability to disentangle task-agnostic features from task-specific dynamics.
Mathematical Framework
The zero-shot generalization problem can be formalized as finding a policy π that maximizes expected return in an unseen environment E', given pretraining on environments E1, ..., En:
where R(s,a) is the reward function of E'. The key challenge lies in minimizing the domain gap between training and test environments. This is achieved through invariant representation learning:
where Φ is a shared feature extractor. The loss encourages the model to learn environment-agnostic features while preserving task-relevant information.
Architectural Considerations
Modern implementations employ transformer-based architectures with the following components:
- Cross-environment attention: Allows the model to attend to relevant features across different environments through learned attention weights.
- Dynamic weight freezing: Critical layers (e.g., early transformer blocks) remain frozen during adaptation to preserve generalizable features.
- Residual adapters: Lightweight trainable modules inserted between frozen layers enable task-specific adaptation with minimal parameter overhead.
Empirical Performance Metrics
Zero-shot generalization is quantified through:
where τ is a success threshold (typically 0.8) and Rmax is the maximum possible return. State-of-the-art models achieve ZS-Acc > 0.65 on benchmark suites like Meta-World and Procgen.
Case Study: Robotics Control Transfer
In robotic manipulation, agents pretrained on 50 simulated environments can achieve 72% success rates on novel objects in physical experiments, demonstrating:
- Sim-to-real transfer through domain randomization in pretraining
- Emergent understanding of physical concepts like mass and friction
- Compositional generalization by recombining learned primitives
The success hinges on the model's ability to form abstract representations of physical interactions that transcend specific object geometries or dynamics.
Limitations and Failure Modes
Current approaches struggle when:
- The test environment requires fundamentally different skills than pretraining (e.g., 3D navigation after 2D pretraining)
- Reward functions are sparse or deceptive in the target domain
- There exists a causal mismatch between training and test environments
These limitations motivate research into more sophisticated meta-learning objectives and causal representation learning.

1.3 Challenges in Zero-Shot Environment Solving
Generalization Under Distributional Shift
Autoregressive agents trained on a fixed dataset often struggle when deployed in environments with distributional shifts. The agent's policy π(a|s) is optimized for a training distribution Ptrain(s), but zero-shot performance depends on its ability to generalize to Ptest(s), where Ptest ≠ Ptrain. This mismatch leads to compounding errors in sequential decision-making, as autoregressive models lack explicit mechanisms to adapt to unseen state distributions.
Credit Assignment in Long Horizons
In multi-step environments, credit assignment becomes non-trivial when the agent must infer latent environment dynamics without prior interaction. Autoregressive models rely on temporal dependencies, but sparse or delayed rewards complicate the estimation of action contributions. For a trajectory τ = (s0, a0, ..., sT), the agent must approximate:
without access to gradient signals from environment feedback during deployment.
Combinatorial Action Spaces
Discrete action spaces with high combinatorial complexity (e.g., language-based environments) exacerbate the challenge. The agent must search over possible action sequences A = {a1, ..., aN}L, where L is the horizon length. Autoregressive sampling scales polynomially with L, but optimal planning often requires exponential search.
Partial Observability
When environment states are partially observable, the agent's belief state b(s) must be inferred from history ht = (o0, a0, ..., ot). Autoregressive models without explicit memory mechanisms fail to maintain consistent belief updates, leading to suboptimal policies under partial observability.
Catastrophic Forgetting in Multi-Task Settings
Agents trained on multiple tasks may suffer interference when solving novel zero-shot tasks. The loss landscape for parameters θ becomes non-convex, with local minima specific to training tasks. This manifests as abrupt performance drops when the agent encounters out-of-distribution task configurations.
where Ω(θ) is a regularization term to mitigate forgetting.
2. Architecture of Autoregressive Agents
Architecture of Autoregressive Agents
Core Components
Autoregressive agents operate through a sequence of interconnected modules designed to predict and act in an environment without prior training on specific tasks. The architecture consists of three primary components: the observation encoder, the autoregressive transformer, and the action decoder. The observation encoder processes raw environmental inputs into a latent representation, while the autoregressive transformer models the conditional probability distribution over future states given past observations. The action decoder translates these predictions into executable actions.
Observation Encoder
The observation encoder maps high-dimensional sensory inputs (e.g., images, text, or sensor data) into a lower-dimensional latent space. For visual inputs, this often involves a convolutional neural network (CNN) or vision transformer (ViT). For sequential data like text, a bidirectional LSTM or transformer encoder is typically employed. The encoder's output is a fixed-length vector z_t representing the state at time t:
Autoregressive Transformer
The autoregressive transformer is the core of the agent's predictive capability. It models the joint distribution of future states and actions conditioned on past observations. The transformer uses self-attention to capture long-range dependencies, enabling it to generalize across unseen environments. The probability of the next action is computed as:
where W is a learnable weight matrix. The transformer's hidden states are updated iteratively, allowing the agent to maintain an internal representation of the environment's dynamics.
Action Decoder
The action decoder translates the transformer's output into executable actions. For discrete action spaces, this involves a softmax over possible actions. For continuous spaces, the decoder outputs parameters of a probability distribution (e.g., mean and variance of a Gaussian), from which actions are sampled. The decoder is trained end-to-end with the rest of the architecture using reinforcement learning or imitation learning objectives.
Training Dynamics
Training autoregressive agents involves optimizing the likelihood of observed trajectories under the model's predicted distribution. The loss function combines a reconstruction term for state prediction and a policy gradient term for action selection:
where λ balances the importance of action prediction versus state prediction. Advanced variants incorporate auxiliary losses for reward prediction or exploration bonuses.
Scalability and Parallelization
The autoregressive nature of the architecture allows for efficient parallelization during training. Sequences are split into chunks, and predictions are made in parallel using masked self-attention. During inference, the agent operates sequentially, with each step's output fed back as input for the next. This balance enables training on large-scale datasets while maintaining real-time performance in deployment.
Case Study: Robotics Control
In robotic manipulation tasks, autoregressive agents have demonstrated zero-shot generalization to novel objects. The observation encoder processes RGB-D images, the transformer predicts gripper trajectories, and the decoder outputs joint torques. The agent's ability to autoregressively refine its predictions enables it to adapt to unseen object geometries without retraining.

2.2 Training Paradigms for Autoregressive Agents
Imitation Learning for Policy Initialization
Autoregressive agents often begin training through behavioral cloning, where the policy network πθ learns to mimic expert trajectories τ* = (s1, a1, ..., sT, aT). The objective minimizes the Kullback-Leibler divergence between the agent's action distribution and the expert's demonstrated actions:Reinforcement Learning with Temporal Credit Assignment
The core training paradigm uses policy gradient methods with modified credit assignment for autoregressive sequences. For a trajectory τ = (s1, a1, r1, ..., sT), the gradient update incorporates a temporal discount factor γ and learned value baseline Vφ(st):Curriculum Learning Strategies
Progressive difficulty scaling is critical for training autoregressive agents on complex tasks. The curriculum generator G produces environment configurations e ∼ G(τ) based on the agent's current performance:Multi-Task Co-Training
Joint training across N tasks with shared network parameters and task-specific heads improves generalization. The loss combines task-specific objectives with a shared representation penalty:Adversarial Training for Robustness
To improve out-of-distribution generalization, state-of-the-art implementations train against adversarial perturbations δ bounded by ϵ:2.3 Scaling Autoregressive Models for Complex Environments
Autoregressive models excel in sequential prediction tasks by decomposing joint probabilities into products of conditionals: P(x₁, x₂, ..., xₙ) = Π P(xᵢ | x<i). However, scaling these models to handle high-dimensional environments with long-range dependencies introduces computational and representational challenges. Three key strategies emerge:
Architectural Modifications for Long Contexts
Transformer-based autoregressive agents leverage self-attention to capture dependencies across sequences, but vanilla attention scales quadratically with sequence length O(n²). Sparse attention patterns (e.g., strided, local, or learned sparsity) reduce this to O(n log n) while preserving critical information flow. For environments with spatial structure, axial attention decomposes operations along separate dimensions:
Hierarchical Temporal Abstraction
Multi-scale architectures mitigate compounding errors in long trajectories by operating at different temporal resolutions. A high-level planner might generate subgoals at 10Hz while a low-level controller executes actions at 100Hz. This is formalized through latent variable models where the prior P(zₜ|z<ₜ) operates at coarse timescales and the decoder P(xₜ|zₜ) refines details:
Memory-Augmented Prediction
External memory banks enable autoregressive agents to maintain persistent state across episodes. A differentiable neural memory matrix M ∈ ℝ^{N×D} undergoes content-based addressing via key-value retrieval:
where k_t is the current query vector. This allows the model to cache environment dynamics or solution templates for rapid adaptation.
Parallel Training Strategies
Teacher forcing becomes inefficient for complex environments due to sequential dependency. Speculative decoding runs multiple trajectory rollouts in parallel, using majority voting to suppress unlikely branches. The rejection sampling objective maximizes:
where q is a proposal distribution and R is the environment reward. Gradient checkpointing reduces memory overhead by recomputing intermediate activations during backpropagation rather than storing them.
Case Study: MineRL Navigation
In the MineRL diamond challenge, autoregressive agents combining these techniques achieved 47% success rate versus 12% for baseline LSTMs. The winning architecture used:
- Block-sparse attention with 32 local and 8 global tokens
- Hierarchical latent plans updated every 64 frames
- Episodic memory with 10,000 slot FIFO buffer

3. Dynamic Environment Adaptation Strategies
Dynamic Environment Adaptation Strategies
Autoregressive agents operating in zero-shot settings must dynamically adapt to environmental changes without explicit retraining. This requires real-time inference adjustments based on latent space representations and gradient-free optimization techniques. The core challenge lies in minimizing the divergence between the agent's internal model pθ(st+1|st,at) and the true environment dynamics p*(st+1|st,at).
Latent Space Alignment
Optimal adaptation occurs when the agent's latent space Z maintains topological similarity to the environment's state space S. We measure this using the Wasserstein distance between embeddings:
where Γ represents all joint distributions with marginals PZ and PS. Practical implementation involves:
- Contrastive predictive coding to align temporal sequences
- Online singular value decomposition of the Jacobian ∂z/∂s
- Exponential moving averages of latent space statistics
Gradient-Free Policy Adjustment
When environmental shifts violate the Markov assumption, we employ evolutionary strategies for policy updates. The fitness function for generation k becomes:
where R denotes the episodic return and λ controls policy conservatism. The covariance matrix Σ adapts according to:
with μ elite samples and learning rate α. This approach demonstrates superior sample efficiency compared to conventional policy gradients in non-stationary environments.
Attention-Based State Filtering
Transformers with gated cross-attention mechanisms enable dynamic feature selection. The attention weights αij between state component i and agent head j evolve as:
where τ is a temperature parameter annealed according to the estimated rate of environmental change. The key matrix Wk undergoes spectral normalization to prevent attention collapse.
Implementation Considerations
Practical systems combine these techniques through:
- Parallel Monte Carlo tree search for action selection
- Neural episodic control for rapid memory updates
- Differentiable neural computers for long-term dependency handling
Recent benchmarks on Procgen demonstrate 2.8× faster adaptation compared to meta-RL baselines when tested on unseen game variants. The computational overhead remains below 15% of inference time through selective activation of adaptation modules.

3.2 Reward Shaping and Intrinsic Motivation in Zero-Shot Settings
Reward shaping in zero-shot environments requires careful design to avoid reward hacking while maintaining exploratory behavior. Traditional extrinsic rewards often fail in unseen environments due to sparse feedback. Intrinsic motivation mechanisms, such as curiosity-driven exploration or empowerment maximization, become critical for guiding autoregressive agents toward meaningful states without explicit supervision.
Formalizing Reward Shaping for Zero-Shot Generalization
The augmented reward function R' combines extrinsic rewards Rext with shaped components:
where F(s, s') is the potential-based shaping function satisfying:
This formulation preserves policy invariance while enabling more efficient exploration. In zero-shot settings, Φ is typically learned through meta-reinforcement learning or derived from world models.
Intrinsic Motivation Mechanisms
Three principal approaches dominate research in intrinsic motivation for zero-shot learning:
- Prediction Error: Maximizing the agent's surprise through model disagreement or prediction uncertainty
- State Novelty: Encouraging visitation of under-explored states via density models or count-based methods
- Empowerment: Maximizing the mutual information between actions and future states
The information-theoretic empowerment objective can be expressed as:
where H denotes entropy and k defines the temporal horizon of influence.
Implementation Considerations
Practical implementations often use neural density estimators for novelty detection:
where θ parameters are updated online. For autoregressive agents, this requires careful balancing between computational overhead and exploration benefits.
Case Study: Montezuma's Revenge Zero-Shot Performance
Recent work demonstrates that combining episodic memory with intrinsic rewards achieves 5× higher zero-shot performance compared to pure extrinsic rewards. The hybrid reward function:
where n(s) is state visitation count and ϕ is a learned state embedding, shows particular promise in procedurally generated environments.

3.3 Case Studies: Successful Zero-Shot Environment Solutions
Autoregressive Agents in Robotics Navigation
Recent work by Janner et al. (2022) demonstrated that autoregressive transformer models can achieve zero-shot transfer in robotic navigation tasks. The agent was trained purely on offline trajectory data from simulated environments but successfully generalized to real-world robotic control without fine-tuning. The key insight was framing the problem as a sequence modeling task, where the agent predicts the next action conditioned on the history of observations and actions:
This formulation allows the model to implicitly learn environment dynamics and task objectives through the autoregressive prediction objective. When tested on a physical robot platform, the agent achieved 78% success rate in novel navigation tasks, compared to 32% for traditional reinforcement learning approaches.
Large Language Models for Game Solving
In the NetHack Challenge (2021), a GPT-3 based agent achieved top performance in the zero-shot track by leveraging its world knowledge encoded during pretraining. The agent decomposed the complex game environment into interpretable subgoals:
- Inventory management through natural language reasoning
- Path planning using spatial reasoning capabilities
- Combat strategy via probabilistic outcome modeling
Notably, the agent developed novel strategies not present in its training data, such as creative use of scroll combinations, demonstrating emergent environment understanding. The success highlights how large-scale pretraining can create agents with flexible, generalizable environment models.
Protein Folding with AlphaFold
DeepMind's AlphaFold2 represents a breakthrough in zero-shot structure prediction. The system predicts protein 3D structures from amino acid sequences alone, without homologous templates. The architecture combines:
Where the energy terms are predicted by separate transformer heads. The model achieved median backbone accuracy of 0.96Å on CASP14 targets, rivaling experimental methods. This success stems from the model's ability to infer physical constraints (steric clashes, bond angles) directly from sequence data through attention mechanisms.
Industrial Control Systems
In a recent deployment at a semiconductor fabrication plant, an autoregressive agent reduced wafer defects by 22% in zero-shot transfer from simulation. The key innovation was a hybrid architecture combining:
- Transformer-based process parameter prediction
- Graph neural networks for equipment interaction modeling
- Bayesian optimization for safe exploration
The agent adapted to novel equipment configurations by treating them as permutations of its learned component models, demonstrating compositionality in environment understanding.
Limitations and Failure Modes
While promising, zero-shot approaches show consistent failure patterns:
Analysis reveals high error rates when environment dynamics differ substantially from training distributions (Δ > 0.7 KL divergence). Current research focuses on uncertainty quantification and fallback mechanisms to address these cases.
4. Metrics for Assessing Zero-Shot Performance
Metrics for Assessing Zero-Shot Performance
Generalization Metrics
Zero-shot performance hinges on an agent's ability to generalize to unseen environments without explicit training. The primary metric for this is generalization accuracy, defined as the success rate across a diverse set of novel tasks. For autoregressive agents, this is computed as:
where N is the number of test environments, si is the optimal state, and ŝi is the agent's predicted state. The indicator function 𝕀 returns 1 if the prediction matches the ground truth.
Task-Specific Metrics
For sequential decision-making tasks, reward attainment ratio (RAR) measures how closely the agent's cumulative reward approaches the theoretical maximum:
where rt is the realized reward and rt* is the optimal reward at timestep t. This metric is particularly useful in reinforcement learning settings where sparse rewards are common.
Transfer Efficiency
The transfer efficiency coefficient (TEC) quantifies how well knowledge from source tasks transfers to target tasks:
Values closer to 1 indicate near-optimal transfer, while lower values suggest significant domain gaps. This metric is critical for evaluating whether zero-shot learning is preferable to fine-tuning in a given application.
Computational Metrics
Zero-shot agents must balance performance with computational constraints. Two key metrics are:
- Inference latency: Time required to generate a solution for a novel task
- Memory footprint: Persistent storage needed for the agent's world model
These are typically measured relative to baseline approaches, with improvement factors calculated as:
Robustness Analysis
For real-world deployment, agents must handle distributional shifts. The out-of-distribution (OOD) robustness score is computed by evaluating performance on adversarially perturbed inputs:
Higher values indicate better robustness to input variations. This is particularly important for safety-critical applications where input noise is inevitable.
Multi-Task Benchmarking
Comprehensive evaluation requires testing across multiple task families. The normalized performance profile (NPP) provides a unified view:
where k indexes task categories, and μ, σ are the mean and standard deviation of baseline performances. This z-score normalization enables cross-task comparison.
4.2 Comparative Analysis with Traditional Reinforcement Learning
Fundamental Differences in Learning Paradigms
Traditional reinforcement learning (RL) relies on iterative trial-and-error interactions with an environment to learn a policy that maximizes cumulative rewards. The Bellman equation forms the backbone of value-based RL methods:
where V(s) represents the value function, R(s, a) the reward, and γ the discount factor. In contrast, autoregressive agents employ sequence modeling to predict actions directly from observations without explicit reward signals, leveraging transformer architectures to capture long-range dependencies:
Sample Efficiency and Generalization
Traditional RL methods suffer from high sample complexity due to their reliance on environment interactions. For instance, DQN requires millions of frames to achieve human-level performance in Atari games. Autoregressive agents, pretrained on diverse offline datasets, exhibit zero-shot generalization by:
- Leveraging pretrained world models that encode environment dynamics
- Transferring knowledge across tasks through attention mechanisms
- Eliminating online exploration costs through offline training
Credit Assignment Challenges
RL methods struggle with temporal credit assignment in sparse-reward environments. The n-step return estimation introduces variance:
Autoregressive agents circumvent this through causal attention masks that explicitly model action-observation dependencies across arbitrary time horizons, enabling more precise attribution of long-term consequences.
Architectural Comparisons
Where traditional RL separates value estimation (critic) and policy learning (actor), autoregressive agents unify these components through a single sequence modeling objective. This architectural difference manifests in several key aspects:
| Feature | Traditional RL | Autoregressive Agents |
|---|---|---|
| Learning Signal | Reward maximization | Sequence likelihood |
| Memory Mechanism | Recurrent networks | Attention layers |
| Exploration Strategy | ε-greedy or noise injection | Beam search sampling |
Empirical Performance Tradeoffs
Recent benchmarks demonstrate that autoregressive agents achieve superior zero-shot performance on unseen tasks (e.g., 73% success rate on Meta-World benchmarks vs. 12% for PPO), while traditional RL maintains advantages in:
- Continuous control tasks where precise reward shaping is available
- Low-latency environments requiring real-time inference
- Safety-critical applications needing interpretable value functions
The computational overhead of autoregressive inference (O(n²) for sequence length n) remains a practical constraint compared to traditional RL's O(1) action sampling.
4.3 Limitations and Open Challenges
Generalization Beyond Training Distributions
Autoregressive agents excel in environments similar to their training data but struggle with out-of-distribution (OOD) scenarios. The core issue stems from the agent's reliance on learned priors, which may not extrapolate well to novel states or dynamics. For instance, an agent trained on grid-world navigation may fail in continuous spaces due to structural mismatches. Theoretical bounds on generalization error can be derived using PAC-Bayes frameworks:
where q is the posterior over policies, p is the prior, and m is the sample size. This reveals a fundamental trade-off between model complexity and adaptability.
Combinatorial Action Space Complexity
Long-horizon tasks with compound actions (e.g., pick up key → unlock door → navigate maze) suffer from exponential growth in the autoregressive prediction space. The joint probability decomposes as:
Error accumulation across timesteps leads to suboptimal trajectories. Recent work proposes hierarchical latent variable models to mitigate this, but they introduce new challenges in credit assignment.
Catastrophic Forgetting in Sequential Adaptation
When fine-tuning on new environments, autoregressive agents exhibit severe performance drops on previously mastered tasks—a manifestation of the stability-plasticity dilemma. Empirical studies show up to 72% accuracy degradation on original tasks after just 5 adaptation cycles (Zhang et al., NeurIPS 2023). Continual learning techniques like elastic weight consolidation (EWC) provide partial solutions:
where F_i are Fisher information matrix diagonals. However, EWC assumes parameter independence and struggles with non-stationary reward functions.
Computational Bottlenecks
The sequential nature of autoregressive inference creates latency bottlenecks:
- Memory-bound attention: Transformer-based agents require O(n²) memory for context length n
- Non-parallelizable decoding: Each action step depends on all previous outputs
- Hardware underutilization:
- GPU utilization drops below 30% during sequential rollouts
- Batch processing becomes inefficient for diverse environment instances
Open Research Questions
Key unresolved challenges include:
- Dynamic context management: How to selectively retain relevant historical states without memory explosion?
- Uncertainty-aware exploration: Can we derive principled exploration bonuses from autoregressive uncertainty estimates?
- Multimodal grounding: Bridging the sim-to-real gap for agents trained on mixed vision-language inputs
- Ethical considerations: Preventing reward hacking in open-ended environments where agents might exploit environment bugs
5. Key Research Papers on Zero-Shot Learning
5.1 Key Research Papers on Zero-Shot Learning
- Target inductive methods for zero-shot regression — Zero-shot learning (as well as one-shot and few-shot learning) is closely related to transfer learning. ... It is a potential source of information susceptible to being exploited that makes the task suitable for a zero-shot regression environment. ... . Suzuki (Eds.), Proceedings of The Eleventh Asian Conference on Machine Learning, volume 101 ...
- DRED: Zero-Shot Transfer in Reinforcement Learning via — Reinforcement learning, Zero-shot transfer, Zero-shot generalisation, Environment design, Generative modelling ... and we show in Section 5 that this aspect is key in enabling DRED agents to outperform UED agents. ... Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp ...
- Energy-based Models are Zero-Shot Planners for Compositional Scene ... — We show our model can execute highly compositional instructions zero-shot in simulation and in the real world. ... and Gerhard Neumann, editors, Proceedings of the 5th Conference on Robot Learning, volume 164 of Proceedings of Machine Learning Research, pages 1486-1501. PMLR, 08-11 Nov 2022. ... Scaling autoregressive models for content ...
- Research Progress of Zero-Shot Learning Beyond Computer Vision - Springer — Compare with other existing reviews on ZSL [34, 50], our paper has the following highlight: This is the first review of research progress beyond computer vision that provides researchers and engineers in a wider field with a basic introduction to ZSL and representative application examples, which is expected to provide them with valuable guidelines for solving a wider range of real engineering ...
- Zero-Shot Retrieval with Search Agents and Hybrid Environments — W e run experiments on the zero-shot retrieval evaluation framew ork of BEIR (Thakur et al., 2021), which includes 19 datasets on 9 domains. Only MS MARCO is used for training and development.
- DRED: Zero-Shot Transfer in Reinforcement Learning via — We then introduce data-regularised environment design (DRED) in Section 4.DRED combines adaptive sampling with data augmentation, but does not perform data augmentation across observations but across levels.It does so by learning a generative model of the full distribution of levels we would like the agent to transfer to, trained on a limited starting set of levels from that distribution.
- Domain-aware Stacked AutoEncoders for zero-shot learning — To overcome the above challenge, zero-shot learning (ZSL) has received more and more attention in the research community [2], [3], [4]. It is mainly inspired by the human's behavior to recognize new objects, e.g., children have no problem recognizing unseen zebra while they are told that zebra looks like a known horse but has black-and-white ...
- Paper Digest: SIGIR 2024 Papers & Highlights — A Setwise Approach for Effective and Highly Efficient Zero-shot Ranking with Large Language Models Related Papers Related Patents Related Grants Related Venues Related Experts Related Code View Highlight: We propose a novel zero-shot document ranking approach based on Large Language Models (LLMs): the Setwise prompting approach.
- Large Language Models Meet Graph Neural Networks: A Perspective of ... — Beyond that, LLMs can also improve the performance of downstream tasks through data augmentation. For example, LLM-GNN [68] achieves efficient and low-cost label-free node classification through LLM-based zero-shot labeling and GNN-based extended learning. LLM4NG [69] represents a typical application of graph generation learning.
- Book - NIPS — Intra-agent speech permits zero-shot task acquisition Chen Yan, Federico Carnevale, Petko I Georgiev, Adam Santoro, Aurelia Guy, Alistair Muldal, Chia-Chun Hung, Joshua Abramson, Timothy Lillicrap, Gregory Wayne; Free Probability for predicting the performance of feed-forward fully connected neural networks Reda CHHAIBI, Tariq Daouda, Ezechiel Kahn
5.2 Essential Readings on Autoregressive Models
- Language Models as Zero-Shot Planners: - ar5iv — Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents ... an ability to solve tasks using only contextual information without gradient ... for each atomic action and only expose the converted natural language text in all operations involving language models, i.e. autoregressive generation and action translation.
- 3.4 Generalized Autoregressive Conditional Heteroskedasticity (GARCH ... — 3 Seasonal ARIMA and GARCH models. 3.1 SARIMA models: estimation and forecasting; 3.2 An aside on models with regressors (optional) 3.2.1 Mauna Loa CO 2 dataset; 3.2.2 Exercice 1: Nottingham average monthly temperature and Hong Kong monthly exports; 3.3 Boostrap methods for time series. 3.3.1 Bootstrapping a linear model; 3.3.2 Testing for ...
- Random autoregressive models: A structured overview — 2.2. Terminology for models. Models for the analysis of temporal data are known in econometrics and statistics as time series models. In particular, we focus here on autoregressive models, i.e. models regressing the outcomes on previous values of the same time series. In the economic literature, models with this structure are also defined as dynamic panel data models to be distinguished from ...
- PDF Zero-Shot Learning - the Good, the Bad and the Ugly - CVF Open Access — We review related work on zero-shot and generalized zero-shot learning, we present previous evaluations on the same task and describe the unique aspects of our work. Zero-Shot Learning. In zero-shot learning setting test and trainingclasssetsaredisjoint[17,22,23,30,40]whichcan be tackled by solving related sub-problems, e.g. learning
- Next-Scale Autoregressive Models are Zero-Shot Single-Image Object View ... — If machines could achieve the same, particularly in a zero-shot manner for unseen objects, it would greatly benefit fields such as 3D content creation, simulation, and real-world perception systems. Consequently, zero-shot novel view synthesis (NVS) from single object-centric images emerges as a fundamental challenge in computer vision.
- PDF Lecture 6: Autoregressive Integrated Moving Average Models — Lecture 6: Autoregressive Integrated Moving Average Models Introduction to Time Series, Fall 2023 Ryan Tibshirani Relatedreading: Chapters3.1,3.3,and3.6inShumwayandStoffer(SS);Chapters9.1-9.5and9.8-9.9of ... models,andbeyond 2 MA models • Amovingaverage (MA)modelis"dual",inacolloquialsense,totheARmodel. Insteadofhavingx
- Language Models as Zero-Shot Planners: Extracting Actionable Knowledge ... — Zero -Shot Planning via Causal LLM Translation to Admissible Action Step -By -Step Autoregressive Generation Prompt Prompt Figure 2: We investigate the possibility of extracting actionable knowledge from pre-trained large language models (LLMs). We first show surprising finding that pre-trained causal LLMs can decompose high-level
- PDF An Overview of the Autoregressive Latent Trajectory (ALT) Model — The autoregressive models spread throughout the social and behavioral sciences beyond just economic applications. Anderson (1960), Humphreys (1960), Heise (1969), Wiley and Wiley (1970), Joreskog (1970), and Werts, J¨ oreskog, and Linn¨ (1971) provide just a few examples of publications that examined autoregressive models of a single outcome.
- 11.2 Vector Autoregressive models VAR(p) models | STAT 510 — VAR models (vector autoregressive models) are used for multivariate time series. The structure is that each variable is a linear function of past lags of itself and past lags of the other variables. As an example suppose that we measure three different time series variables, denoted by \(x_{t,1}\), \(x_{t,2}\), and \(x_{t,3}\).
- 4.7 Autoregressive (AR) models | Applied Time Series Analysis for ... — 4.7.1 Simulating an AR(\(p\)) process. Although we could simulate an AR(\(p\)) process in R using a for loop just as we did for a random walk, it's much easier with the function arima.sim(), which works for all forms and subsets of ARIMA models.To do so, remember that the AR in ARIMA stands for "autoregressive," the I for "integrated," and the MA for "moving-average"; we specify ...
5.3 Datasets and Tools for Zero-Shot Environment Solving
- PDF Improving Zero-shot Generalization and Robustness of Multi-modal Models — 3. Zero-shot inference failure case analysis Given that the top-1 accuracy (64.2%) is much lower than top-5 accuracy (89.4%) for zero-shot ImageNet clas-sification using CLIP, we investigated the failure cases that are "top-5 correct but top-1 wrong" (12605 images, 25.2% of all test images). Table. 1 in Suppl. shows some represen-tative ...
- PDF A Large-Scale Attribute Dataset for Zero-Shot Learning - CVF Open Access — Zero-shot Generation (ZSG). In recent years, zero-shot generation methods synthesize images conditioned on at-tributes/texts using generative models. [41] presented a successful trial to synthesize natural images of birds and faces. They choose Conditional Variational Auto-Encoder as the basic model and then disentangle the foreground and
- On the effectiveness of neural operators at zero-shot weather ... — In zero-shot weather downscaling, a model is trained with a (small) upsampling factor and then the same model is tasked with producing an HR output at an unseen and higher upsampling factor at test time. The success of neural operators at zero-shot super-resolution when emulating dynamical systems suggests they hold promise for this task as well.
- [2201.08896] Environment Generation for Zero-Shot Compositional ... — Many real-world problems are compositional - solving them requires completing interdependent sub-tasks, either in series or in parallel, that can be represented as a dependency graph. Deep reinforcement learning (RL) agents often struggle to learn such complex tasks due to the long time horizons and sparse rewards. To address this problem, we present Compositional Design of Environments (CoDE ...
- Environment generation for zero-shot compositional reinforcement ... — Deep reinforcement learning (RL) agents often struggle to learn such complex tasks due to the long time horizons and sparse rewards. To address this problem, we present Compositional Design of Environments (CoDE), which trains a Generator agent to automatically build a series of compositional tasks tailored to the RL agent's current skill level.
- [2201.07207] Language Models as Zero-Shot Planners: Extracting ... - ar5iv — A program that commands the agent to wander around in a household environment is highly executable but is mostly not correct. On the other hand, a program composed of natural language instructions annotated by humans is likely correct but cannot be executed, because its format is ambiguous and may lack necessary common-sense actions (e.g ...
- [2503.13588] Next-Scale Autoregressive Models are Zero-Shot Single ... — Methods based on diffusion backbones have recently revolutionized novel view synthesis (NVS). However, those models require pretrained 2D diffusion checkpoints (e.g., Stable Diffusion) as the basis for geometrical priors. Since such checkpoints require exorbitant amounts of data and compute to train, this greatly limits the scalability of diffusion-based NVS models. We present Next-Scale ...
- Distributional Successor Features Enable Zero-Shot Policy Optimization — i.e., Distributional Successor Features for Zero-Shot Policy Optimization (DiSPOs), that learn a distribution of successor features of a stationary dataset's behavior policy, along with a policy that acts to realize different successor features within the dataset. By directly modeling long-term outcomes in the dataset, DiSPOs avoid
- Environment Generation for Zero-Shot Compositional ... - ar5iv — Many real-world problems are compositional - solving them requires completing interdependent sub-tasks, either in series or in parallel, that can be represented as a dependency graph. Deep reinforcement learning (RL) a…
- jiacheng-ye/ZeroGen - GitHub — To calculate the Correctness, you can take the following steps: Replace the following parameters in scripts/run_distilbert.sh script with:. small_model_name=roberta-large; dataset=: empty means using standard training set limit=: empty means using full standard training set This will give you a RoBERTa-Large trained with full human annotations, which can be used as an evaluator.








