We propose a set of modifications that improve training time and likelihood estimation of hierarchical mixture models implemented via Probabilistic Circuits (PCs). Our proposal reduces the complexity of mutual information estimation in the structure learning step of PCs from quadratic to linear in the number of inputs, without sacrificing likelihood estimation performance on image datasets. We repurpose invertible transformations from the lossless compression community to improve likelihood estimation by a factor of up to 25% on benchmark image datasets, making PCs competitive with current standard codecs on low-resolution datasets. Despite our improvements, experiments with low- and high-resolution image datasets indicate that the advantage of lossless neural compression and PCs over standard codecs, such as WebP, disappears as the image size increases, motivating future work on practical lossless neural compression.

Data compression is a critical component of today's digital infrastructure, in particular for storage and transmission of visual data over the internet. While end-to-end neural codecs have been shown to outperform hand-designed image and video codecs, their mainstream adoption is hindered by their high computational cost. In this paper we reassess standard design choices for neural image compression autoencoders, identifying two inefficiencies: the first is the substantial compute spent on high-resolution feature maps, and the second is methods of applying activations that lead to less expressive representations. We mitigate both issues and propose a new architecture, PatchMixer, that begins with approximately patchwise-independent encoding, followed by mixing layers, thus enabling compute savings. PatchMixer achieves a BD-rate within 5% of VVC on Kodak with 97 kMACs per pixel of decoding complexity and 200 kMACs per pixel of cumulative complexity, to our knowledge the lowest cumulative complexity for a neural codec within 5% of VVC.

Recent advances in text-to-image (T2I) models have achieved impressive quality and consistency. However, this has come at the cost of representation diversity. While automatic evaluation methods exist for benchmarking model diversity, they either require reference image datasets or lack specificity about the kind of diversity measured, limiting their adaptability and interpretability. To address this gap, we introduce the Does-it/Can-it framework, DIM-CIM, a reference-free measurement of default-mode diversity ("Does" the model generate images with expected attributes?) and generalization capacity ("Can" the model generate diverse attributes for a particular concept?). We construct the COCO-DIMCIM benchmark, which is seeded with COCO concepts and captions and augmented by a large language model. With COCO-DIMCIM, we find that widely-used models improve in generalization at the cost of default-mode diversity when scaling from 1.5B to 8.1B parameters. DIMCIM also identifies fine-grained failure cases, such as attributes that are generated with generic prompts but are rarely generated when explicitly requested. Finally, we use DIMCIM to evaluate the training data of a T2I model and observe a correlation of 0.85 between diversity in training images and default-mode diversity. Our work provides a flexible and interpretable framework for assessing T2I model diversity and generalization, enabling a more comprehensive understanding of model performance.

Tokenization is associated with many poorly understood shortcomings in language models (LMs), yet remains an important component for long sequence scaling purposes. This work studies how tokenization impacts model performance by analyzing and comparing the stochastic behavior of tokenized models with their byte-level, or token-free, counterparts. We discover that, even when the two models are statistically equivalent, their predictive distributions over the next byte can be substantially different, a phenomenon we term as ``tokenization bias''. To fully characterize this phenomenon, we introduce the Byte-Token Representation Lemma, a framework that establishes a mapping between the learned token distribution and its equivalent byte-level distribution. From this result, we develop a next-byte sampling algorithm that eliminates tokenization bias without requiring further training or optimization. In other words, this enables zero-shot conversion of tokenized LMs into statistically equivalent token-free ones. We demonstrate its broad applicability with two use cases: fill-in-the-middle (FIM) tasks and model ensembles. In FIM tasks where input prompts may terminate mid-token, leading to out-of-distribution tokenization, our method mitigates performance degradation and achieves 18\% improvement in FIM coding benchmarks, while consistently outperforming the standard token healing fix. For model ensembles where each model employs a distinct vocabulary, our approach enables seamless integration, resulting in improved performance up to 3.7\% over individual models across various standard baselines in reasoning, knowledge, and coding. Code is available at:https: //github.com/facebookresearch/Exact-Byte-Level-Probabilities-from-Tokenized-LMs.

To date, the multi-objective optimization literature has mainly focused on conflicting objectives, studying the Pareto front, or requiring users to balance tradeoffs. Yet, in machine learning practice, there are many scenarios where such conflict does not take place. Recent findings from multi-task learning, reinforcement learning, and LLMs training show that diverse related tasks can enhance performance across objectives simultaneously. Despite this evidence, such phenomenon has not been examined from an optimization perspective. This leads to a lack of generic gradient-based methods that can scale to scenarios with a large number of related objectives. To address this gap, we introduce the Aligned Multi-Objective Optimization framework, propose new algorithms for this setting, and provide theoretical guarantees of its superior performance compared to naive approaches.

Knowing the effect of an intervention is critical for human decision-making, but current approaches for causal effect estimation rely on manual data collection and structuring, regardless of the causal assumptions. This increases both the cost and time-to-completion for studies. We show how large, diverse observational text data can be mined with large language models (LLMs) to produce inexpensive causal effect estimates under appropriate causal assumptions. We introduce _NATURAL_, a novel family of causal effect estimators built with LLMs that operate over datasets of unstructured text. Our estimators use LLM conditional distributions (over variables of interest, given the text data) to assist in the computation of classical estimators of causal effect. We overcome a number of technical challenges to realize this idea, such as automating data curation and using LLMs to impute missing information. We prepare six (two synthetic and four real) observational datasets, paired with corresponding ground truth in the form of randomized trials, which we used to systematically evaluate each step of our pipeline. NATURAL estimators demonstrate remarkable performance, yielding causal effect estimates that fall within 3 percentage points of their ground truth counterparts, including on real-world Phase 3/4 clinical trials. Our results suggest that unstructured text data is a rich source of causal effect information, and NATURAL is a first step towards an automated pipeline to tap this resource.

Large language models (LLMs) are trained on a deluge of text data with limited quality control. As a result, LLMs can exhibit unintended or even harmful behaviours, such as leaking information, fake news or hate speech. Countermeasures, commonly referred to as preference alignment, include fine-tuning the pretrained LLMs with carefully crafted text examples of desired behaviour. Even then, empirical evidence shows preference aligned LLMs can be enticed to harmful behaviour. This so called jailbreaking of LLMs is typically achieved by adversarially modifying the input prompt to the LLM. Our paper provides theoretical insights into the phenomenon of preference alignment and jailbreaking from a statistical perspective. Under our framework, we first show that pretrained LLMs will mimic harmful behaviour if present in the training corpus. \textbf{Under that same framework, we then introduce a statistical notion of alignment, and lower-bound the jailbreaking probability, showing that it is unpreventable under reasonable assumptions.} Based on our insights, we propose an alteration to the currently prevalent alignment strategy RLHF. Specifically, we introduce a simple modification to the RLHF objective, we call \emph{E-RLHF}, that aims to increase the likelihood of safe responses. \emph{E-RLHF} brings no additional training cost, and is compatible with other methods. Empirically, we demonstrate that \emph{E-RLHF} outperforms RLHF on all alignment problems put forward by the AdvBench \citep{zou2023universal} and HarmBench project \citep{mazeika2024harmbench} without sacrificing model performance as measured by the MT-Bench project \citep{zheng2024judging}.

Lossy image compression aims to represent images in as few bits as possible while maintaining fidelity to the original. Theoretical results indicate that optimizing distortion metrics such as PSNR or MS-SSIM necessarily leads to a discrepancy in the statistics of original images from those of reconstructions, in particular at low bitrates, often manifested by the blurring of the compressed images. Previous work has leveraged adversarial discriminators to improve statistical fidelity. Yet these binary discriminators adopted from generative modeling tasks may not be ideal for image compression. In this paper, we introduce a non-binary discriminator that is conditioned on quantized local image representations obtained via VQ-VAE autoencoders. Our evaluations on the CLIC2020, DIV2K and Kodak datasets show that our discriminator is more effective for jointly optimizing distortion (e.g., PSNR) and statistical fidelity (e.g., FID) than the PatchGAN of the state-of-the-art HiFiC model. On CLIC2020, we obtain the same FID as HiFiC with 30-40% fewer bits.

Current methods which compress multisets at an optimal rate have computational complexity that scales linearly with alphabet size, making them too slow to be practical in many real-world settings. We show how to convert a compression algorithm for sequences into one for multisets, in exchange for an additional complexity term that is quasi-linear in sequence length. This allows us to compress multisets of independent and identically distributed symbols at an optimal rate, with computational complexity decoupled from the alphabet size. The key insight is to avoid encoding the multiset directly, and instead compress a proxy sequence, using a technique called ‘bits-back coding’. We demonstrate the method experimentally on two tasks which are intractible with previous optimal-rate methods: compression of multisets of images and JavaScript Object Notation (JSON) files. Code for our experiments is available at https://github.com/facebookresearch/multiset-compression.

Latent variable models have been successfully applied in lossless compression with the bits-back coding algorithm. However, bits-back suffers from an increase in the bitrate equal to the KL divergence between the approximate posterior and the true posterior. In this paper, we show how to remove this gap asymptotically by deriving bits-back coding algorithms from tighter variational bounds. The key idea is to exploit extended space representations of Monte Carlo estimators of the marginal likelihood. Naively applied, our schemes would require more initial bits than the standard bits-back coder, but we show how to drastically reduce this additional cost with couplings in the latent space. When parallel architectures can be exploited, our coders can achieve better rates than bits-back with little additional cost. We demonstrate improved lossless compression rates in a variety of settings, especially in out-of-distribution or sequential data compression.

Most data is automatically collected and only ever "seen" by algorithms. Yet, data compressors preserve perceptual fidelity rather than just the information needed by algorithms performing downstream tasks. In this paper, we characterize the bit-rate required to ensure high performance on all predictive tasks that are invariant under a set of transformations, such as data augmentations. Based on our theory, we design unsupervised objectives for training neural compressors. Using these objectives, we train a generic image compressor that achieves substantial rate savings (more than 1000x on ImageNet) compared to JPEG on 8 datasets, without decreasing downstream classification performance.

Compression and computational efficiency in deep learning have become a problem of great significance. In this work, we argue that the most principled and effective way to attack this problem is by adopting a Bayesian point of view, where through sparsity inducing priors we prune large parts of the network. We introduce two novelties in this paper: 1) we use hierarchical priors to prune nodes instead of individual weights, and 2) we use the posterior uncertainties to determine the optimal fixed point precision to encode the weights. Both factors significantly contribute to achieving the state of the art in terms of compression rates, while still staying competitive with methods designed to optimize for speed or energy efficiency.