LaFFi: Leveraging Hybrid Natural Language Feedback for Fine-tuning Language Models

Fine-tuning Large Language Models (LLMs) adapts a trained model to specific downstream tasks, significantly improving task-specific performance. Supervised Fine-Tuning (SFT) is a common approach, where an LLM is trained to produce desired answers. However, LLMs trained with SFT sometimes make simple mistakes and result in hallucinations on reasoning tasks such as question-answering. Without external feedback, it is difficult for SFT to learn a good mapping between the question and the desired answer, especially with a small dataset. This paper introduces an alternative to SFT called Natural Language Feedback for Finetuning LLMs (LaFFi). LaFFi has LLMs directly predict the feedback they will receive from an annotator. We find that requiring such reflection can significantly improve the accuracy in in-domain question-answering tasks, providing a promising direction for the application of natural language feedback in the realm of SFT LLMs. Additional ablation studies show that the portion of human-annotated data in the annotated datasets affects the fine-tuning performance.

EmbodiedGPT: Vision-Language Pre-Training via Embodied Chain of Thought

Embodied AI is a crucial frontier in robotics, capable of planning and executing action sequences for robots to accomplish long-horizon tasks in physical environments. In this work, we introduce EmbodiedGPT, an end-to-end multi-modal foundation model for embodied AI, empowering embodied agents with multi-modal understanding and execution capabilities. To achieve this, we have made the following efforts: (i) We craft a large-scale embodied planning dataset, termed EgoCOT. The dataset consists of carefully selected videos from the Ego4D dataset, along with corresponding high-quality language instructions. Specifically, we generate a sequence of sub-goals with the "Chain of Thoughts" mode for effective embodied planning. (ii) We introduce an efficient training approach to EmbodiedGPT for high-quality plan generation, by adapting a 7B large language model (LLM) to the EgoCOT dataset via prefix tuning. (iii) We introduce a paradigm for extracting task-related features from LLM-generated planning queries to form a closed loop between high-level planning and low-level control. Extensive experiments show the effectiveness of EmbodiedGPT on embodied tasks, including embodied planning, embodied control, visual captioning, and visual question answering. Notably, EmbodiedGPT significantly enhances the success rate of the embodied control task by extracting more effective features. It has achieved a remarkable 1.6 times increase in success rate on the Franka Kitchen benchmark and a 1.3 times increase on the Meta-World benchmark, compared to the BLIP-2 baseline fine-tuned with the Ego4D dataset.

A Simple Decentralized Cross-Entropy Method

Cross-Entropy Method (CEM) is commonly used for planning in model-based reinforcement learning (MBRL) where a centralized approach is typically utilized to update the sampling distribution based on only the top-$k$ operation's results on samples. In this paper, we show that such a centralized approach makes CEM vulnerable to local optima, thus impairing its sample efficiency. To tackle this issue, we propose Decentralized CEM (DecentCEM), a simple but effective improvement over classical CEM, by using an ensemble of CEM instances running independently from one another, and each performing a local improvement of its own sampling distribution. We provide both theoretical and empirical analysis to demonstrate the effectiveness of this simple decentralized approach. We empirically show that, compared to the classical centralized approach using either a single or even a mixture of Gaussian distributions, our DecentCEM finds the global optimum much more consistently thus improves the sample efficiency. Furthermore, we plug in our DecentCEM in the planning problem of MBRL, and evaluate our approach in several continuous control environments, with comparison to the state-of-art CEM based MBRL approaches (PETS and POPLIN). Results show sample efficiency improvement by simply replacing the classical CEM module with our DecentCEM module, while only sacrificing a reasonable amount of computational cost. Lastly, we conduct ablation studies for more in-depth analysis. Code is available at https://github.com/vincentzhang/decentCEM

Variable-Decision Frequency Option Critic

In classic reinforcement learning algorithms, agents make decisions at discrete and fixed time intervals. The physical duration between one decision and the next becomes a critical hyperparameter. When this duration is too short, the agent needs to make many decisions to achieve its goal, aggravating the problem's difficulty. But when this duration is too long, the agent becomes incapable of controlling the system. Physical systems, however, do not need a constant control frequency. For learning agents, it is desirable to operate with low frequency when possible and high frequency when necessary. We propose a framework called Continuous-Time Continuous-Options (CTCO), where the agent chooses options as sub-policies of variable durations. Such options are time-continuous and can interact with the system at any desired frequency providing a smooth change of actions. The empirical analysis shows that our algorithm is competitive w.r.t. other time-abstraction techniques, such as classic option learning and action repetition, and practically overcomes the difficult choice of the decision frequency.

CWD: A Machine Learning based Approach to Detect Unknown Cloud Workloads

Workloads in modern cloud data centers are becoming increasingly complex. The number of workloads running in cloud data centers has been growing exponentially for the last few years, and cloud service providers (CSP) have been supporting on-demand services in real-time. Realizing the growing complexity of cloud environment and cloud workloads, hardware vendors such as Intel and AMD are increasingly introducing cloud-specific workload acceleration features in their CPU platforms. These features are typically targeted towards popular and commonly-used cloud workloads. Nonetheless, uncommon, customer-specific workloads (unknown workloads), if their characteristics are different from common workloads (known workloads), may not realize the potential of the underlying platform. To address this problem of realizing the full potential of the underlying platform, we develop a machine learning based technique to characterize, profile and predict workloads running in the cloud environment. Experimental evaluation of our technique demonstrates good prediction performance. We also develop techniques to analyze the performance of the model in a standalone manner.

Build generally reusable agent-environment interaction models

This paper tackles the problem of how to pre-train a model and make it generally reusable backbones for downstream task learning. In pre-training, we propose a method that builds an agent-environment interaction model by learning domain invariant successor features from the agent's vast experiences covering various tasks, then discretize them into behavior prototypes which result in an embodied set structure. To make the model generally reusable for downstream task learning, we propose (1) embodied feature projection that retains previous knowledge by projecting the new task's observation-action pair to the embodied set structure and (2) projected Bellman updates which add learning plasticity for the new task setting. We provide preliminary results that show downstream task learning based on a pre-trained embodied set structure can handle unseen changes in task objectives, environmental dynamics and sensor modalities.

Auxiliary task discovery through generate-and-test

In this paper, we explore an approach to auxiliary task discovery in reinforcement learning based on ideas from representation learning. Auxiliary tasks tend to improve data efficiency by forcing the agent to learn auxiliary prediction and control objectives in addition to the main task of maximizing reward, and thus producing better representations. Typically these tasks are designed by people. Meta-learning offers a promising avenue for automatic task discovery; however, these methods are computationally expensive and challenging to tune in practice. In this paper, we explore a complementary approach to the auxiliary task discovery: continually generating new auxiliary tasks and preserving only those with high utility. We also introduce a new measure of auxiliary tasks usefulness based on how useful the features induced by them are for the main task. Our discovery algorithm significantly outperforms random tasks, hand-designed tasks, and learning without auxiliary tasks across a suite of environments.

What makes useful auxiliary tasks in reinforcement learning: investigating the effect of the target policy

Auxiliary tasks have been argued to be useful for representation learning in reinforcement learning. Although many auxiliary tasks have been empirically shown to be effective for accelerating learning on the main task, it is not yet clear what makes useful auxiliary tasks. Some of the most promising results are on the pixel control, reward prediction, and the next state prediction auxiliary tasks; however, the empirical results are mixed, showing substantial improvements in some cases and marginal improvements in others. Careful investigations of how auxiliary tasks help the learning of the main task is necessary. In this paper, we take a step studying the effect of the target policies on the usefulness of the auxiliary tasks formulated as general value functions. General value functions consist of three core elements: 1) policy 2) cumulant 3) continuation function. Our focus on the role of the target policy of the auxiliary tasks is motivated by the fact that the target policy determines the behavior about which the agent wants to make a prediction and the state-action distribution that the agent is trained on, which further affects the main task learning. Our study provides insights about questions such as: Does a greedy policy result in bigger improvement gains compared to other policies? Is it best to set the auxiliary task policy to be the same as the main task policy? Does the choice of the target policy have a substantial effect on the achieved performance gain or simple strategies for setting the policy, such as using a uniformly random policy, work as well? Our empirical results suggest that: 1) Auxiliary tasks with the greedy policy tend to be useful. 2) Most policies, including a uniformly random policy, tend to improve over the baseline. 3) Surprisingly, the main task policy tends to be less useful compared to other policies.

Generalizable task representation learning from human demonstration videos: a geometric approach

We study the problem of generalizable task learning from human demonstration videos without extra training on the robot or pre-recorded robot motions. Given a set of human demonstration videos showing a task with different objects/tools (categorical objects), we aim to learn a representation of visual observation that generalizes to categorical objects and enables efficient controller design. We propose to introduce a geometric task structure to the representation learning problem that geometrically encodes the task specification from human demonstration videos, and that enables generalization by building task specification correspondence between categorical objects. Specifically, we propose CoVGS-IL, which uses a graph-structured task function to learn task representations under structural constraints. Our method enables task generalization by selecting geometric features from different objects whose inner connection relationships define the same task in geometric constraints. The learned task representation is then transferred to a robot controller using uncalibrated visual servoing (UVS); thus, the need for extra robot training or pre-recorded robot motions is removed.

A Quantitative Analysis of Activities of Daily Living: Insights into Improving Functional Independence with Assistive Robotics

Human assistive robotics have the potential to help the elderly and individuals living with disabilities with their Activities of Daily Living (ADL). Robotics researchers focus on assistive tasks from the perspective of various control schemes and motion types. Health research on the other hand focuses on clinical assessment and rehabilitation, arguably leaving important differences between the two domains. In particular, little is known quantitatively on which ADLs are typically carried out in a persons everyday environment - at home, work, etc. Understanding what activities are frequently carried out during the day can help guide the development and prioritization of robotic technology for in-home assistive robotic deployment. This study targets several lifelogging databases, where we compute (i) ADL task frequency from long-term low sampling frequency video and Internet of Things (IoT) sensor data, and (ii) short term arm and hand movement data from 30 fps video data of domestic tasks. Robotics and health care communities have differing terms and taxonomies for representing tasks and motions. In this work, we derive and discuss a robotics-relevant taxonomy from quantitative ADL task and motion data in attempt to ameliorate taxonomic differences between the two communities. Our quantitative results provide direction for the development of better assistive robots to support the true demands of the healthcare community.