In this paper, we study how to synthesize a dynamic reference from an external dictionary to perform conditional coding of the input image in the latent domain and how to learn the conditional latent synthesis and coding modules in an end-to-end manner. Our approach begins by constructing a universal image feature dictionary using a multi-stage approach involving modified spatial pyramid pooling, dimension reduction, and multi-scale feature clustering. For each input image, we learn to synthesize a conditioning latent by selecting and synthesizing relevant features from the dictionary, which significantly enhances the model's capability in capturing and exploring image source correlation. This conditional latent synthesis involves a correlation-based feature matching and alignment strategy, comprising a Conditional Latent Matching (CLM) module and a Conditional Latent Synthesis (CLS) module. The synthesized latent is then used to guide the encoding process, allowing for more efficient compression by exploiting the correlation between the input image and the reference dictionary. According to our theoretical analysis, the proposed conditional latent coding (CLC) method is robust to perturbations in the external dictionary samples and the selected conditioning latent, with an error bound that scales logarithmically with the dictionary size, ensuring stability even with large and diverse dictionaries. Experimental results on benchmark datasets show that our new method improves the coding performance by a large margin (up to 1.2 dB) with a very small overhead of approximately 0.5\% bits per pixel. Our code is publicly available at https://github.com/ydchen0806/CLC.

A key challenge in interpretability is to decompose model activations into meaningful features. Sparse autoencoders (SAEs) have emerged as a promising tool for this task. However, a central problem in evaluating the quality of SAEs is the absence of ground truth features to serve as an evaluation gold standard. Current evaluation methods for SAEs are therefore confronted with a significant trade-off: SAEs can either leverage toy models or other proxies with predefined ground truth features; or they use extensive prior knowledge of realistic task circuits. The former limits the generalizability of the evaluation results, while the latter limits the range of models and tasks that can be used for evaluations. We introduce SAGE: Scalable Autoencoder Ground-truth Evaluation, a ground truth evaluation framework for SAEs that scales to large state-of-the-art SAEs and models. We demonstrate that our method can automatically identify task-specific activations and compute ground truth features at these points. Compared to previous methods we reduce the training overhead by introducing a novel reconstruction method that allows to apply residual stream SAEs to sublayer activations. This eliminates the need for SAEs trained on every task-specific activation location. Then we validate the scalability of our framework, by evaluating SAEs on novel tasks on Pythia70M, GPT-2 Small, and Gemma-2-2. Our framework therefore paves the way for generalizable, large-scale evaluations of SAEs in interpretability research.

Disentangling model activations into meaningful features is a central problem in interpretability. However, the absence of ground-truth for these features in realistic scenarios makes validating recent approaches, such as sparse dictionary learning, elusive. To address this challenge, we propose a framework for evaluating feature dictionaries in the context of specific tasks, by comparing them against \emph{supervised} feature dictionaries. First, we demonstrate that supervised dictionaries achieve excellent approximation, control, and interpretability of model computations on the task. Second, we use the supervised dictionaries to develop and contextualize evaluations of unsupervised dictionaries along the same three axes. We apply this framework to the indirect object identification (IOI) task using GPT-2 Small, with sparse autoencoders (SAEs) trained on either the IOI or OpenWebText datasets. We find that these SAEs capture interpretable features for the IOI task, but they are less successful than supervised features in controlling the model. Finally, we observe two qualitative phenomena in SAE training: feature occlusion (where a causally relevant concept is robustly overshadowed by even slightly higher-magnitude ones in the learned features), and feature over-splitting (where binary features split into many smaller, less interpretable features). We hope that our framework will provide a useful step towards more objective and grounded evaluations of sparse dictionary learning methods.

In enterprise ML architectures, it's wise to maintain the outputs of the feature jobs in a sharable format without encoding. These features can be later cherrypicked, encoded, and fed into an ML model that needs it. This approach has several advantages. When features are readily available, the journey from a'business question' to'scientific answer' becomes much more simple. With the availability of feature pool, when a data scientist wants to do a new experiment, he/she does not have to start from the raw data. Instead he/she can start with the available features. This can avoid a lot of unoptimised runs. In the cases where they need more data-features, it can go as a request to the engineering team to optimally build whatever new is requested. And when they are confident to take the model to production environment, the model promotion will involve only minimal components.

Feature selection and regularization are becoming increasingly prominent tools in the efforts of the reinforcement learning (RL) community to expand the reach and applicability of RL. One approach to the problem of feature selection is to impose a sparsity-inducing form of regularization on the learning method. Recent work on $L_1$ regularization has adapted techniques from the supervised learning literature for use with RL. Another approach that has received renewed attention in the supervised learning community is that of using a simple algorithm that greedily adds new features. Such algorithms have many of the good properties of the $L_1$ regularization methods, while also being extremely efficient and, in some cases, allowing theoretical guarantees on recovery of the true form of a sparse target function from sampled data. This paper considers variants of orthogonal matching pursuit (OMP) applied to reinforcement learning. The resulting algorithms are analyzed and compared experimentally with existing $L_1$ regularized approaches. We demonstrate that perhaps the most natural scenario in which one might hope to achieve sparse recovery fails; however, one variant, OMP-BRM, provides promising theoretical guarantees under certain assumptions on the feature dictionary. Another variant, OMP-TD, empirically outperforms prior methods both in approximation accuracy and efficiency on several benchmark problems.