Mitigating Embedding Collapse in Diffusion Models for Categorical Data

Latent diffusion models have enabled continuous-state diffusion models to handle a variety of datasets, including categorical data. However, most methods rely on fixed pretrained embeddings, limiting the benefits of joint training with the diffusion model. While jointly learning the embedding (via reconstruction loss) and the latent diffusion model (via score matching loss) could enhance performance, our analysis shows that end-to-end training risks embedding collapse, degrading generation quality. To address this issue, we introduce CATDM, a continuous diffusion framework within the embedding space that stabilizes training. We propose a novel objective combining the joint embedding-diffusion variational lower bound with a Consistency-Matching (CM) regularizer, alongside a shifted cosine noise schedule and random dropping strategy. The CM regularizer ensures the recovery of the true data distribution. Experiments on benchmarks show that CATDM mitigates embedding collapse, yielding superior results on FFHQ, LSUN Churches, and LSUN Bedrooms. In particular, CATDM achieves an FID of 6.81 on ImageNet $256\times256$ with 50 steps. It outperforms non-autoregressive models in machine translation and is on a par with previous methods in text generation.

SAFT: Towards Out-of-Distribution Generalization in Fine-Tuning

Handling distribution shifts from training data, known as out-of-distribution (OOD) generalization, poses a significant challenge in the field of machine learning. While a pre-trained vision-language model like CLIP has demonstrated remarkable zero-shot performance, further adaptation of the model to downstream tasks leads to undesirable degradation for OOD data. In this work, we introduce Sparse Adaptation for Fine-Tuning (SAFT), a method that prevents fine-tuning from forgetting the general knowledge in the pre-trained model. SAFT only updates a small subset of important parameters whose gradient magnitude is large, while keeping the other parameters frozen. SAFT is straightforward to implement and conceptually simple. Extensive experiments show that with only 0.1% of the model parameters, SAFT can significantly improve the performance of CLIP. It consistently outperforms baseline methods across several benchmarks. On the few-shot learning benchmark of ImageNet and its variants, SAFT gives a gain of 5.15% on average over the conventional fine-tuning method in OOD settings.

SPARO: Selective Attention for Robust and Compositional Transformer Encodings for Vision

Selective attention helps us focus on task-relevant aspects in the constant flood of our sensory input. This constraint in our perception allows us to robustly generalize under distractions and to new compositions of perceivable concepts. Transformers employ a similar notion of attention in their architecture, but representation learning models with transformer backbones like CLIP and DINO often fail to demonstrate robustness and compositionality. We highlight a missing architectural prior: unlike human perception, transformer encodings do not separately attend over individual concepts. In response, we propose SPARO, a read-out mechanism that partitions encodings into separately-attended slots, each produced by a single attention head. Using SPARO with CLIP imparts an inductive bias that the vision and text modalities are different views of a shared compositional world with the same corresponding concepts. Using SPARO, we demonstrate improvements on downstream recognition, robustness, retrieval, and compositionality benchmarks with CLIP (up to +14% for ImageNet, +4% for SugarCrepe), and on nearest neighbors and linear probe for ImageNet with DINO (+3% each). We also showcase a powerful ability to intervene and select individual SPARO concepts to further improve downstream task performance (up from +4% to +9% for SugarCrepe) and use this ability to study the robustness of SPARO's representation structure. Finally, we provide insights through ablation experiments and visualization of learned concepts.

SKILL: Similarity-aware Knowledge distILLation for Speech Self-Supervised Learning

Self-supervised learning (SSL) has achieved remarkable success across various speech-processing tasks. To enhance its efficiency, previous works often leverage the use of compression techniques. A notable recent attempt is DPHuBERT, which applies joint knowledge distillation (KD) and structured pruning to learn a significantly smaller SSL model. In this paper, we contribute to this research domain by introducing SKILL, a novel method that conducts distillation across groups of layers instead of distilling individual arbitrarily selected layers within the teacher network. The identification of the layers to distill is achieved through a hierarchical clustering procedure applied to layer similarity measures. Extensive experiments demonstrate that our distilled version of WavLM Base+ not only outperforms DPHuBERT but also achieves state-of-the-art results in the 30M parameters model class across several SUPERB tasks.

Towards Robust FastSpeech 2 by Modelling Residual Multimodality

State-of-the-art non-autoregressive text-to-speech (TTS) models based on FastSpeech 2 can efficiently synthesise high-fidelity and natural speech. For expressive speech datasets however, we observe characteristic audio distortions. We demonstrate that such artefacts are introduced to the vocoder reconstruction by over-smooth mel-spectrogram predictions, which are induced by the choice of mean-squared-error (MSE) loss for training the mel-spectrogram decoder. With MSE loss FastSpeech 2 is limited to learn conditional averages of the training distribution, which might not lie close to a natural sample if the distribution still appears multimodal after all conditioning signals. To alleviate this problem, we introduce TVC-GMM, a mixture model of Trivariate-Chain Gaussian distributions, to model the residual multimodality. TVC-GMM reduces spectrogram smoothness and improves perceptual audio quality in particular for expressive datasets as shown by both objective and subjective evaluation.

Efficient Training of Deep Equilibrium Models

Deep equilibrium models (DEQs) have proven to be very powerful for learning data representations. The idea is to replace traditional (explicit) feedforward neural networks with an implicit fixed-point equation, which allows to decouple the forward and backward passes. In particular, training DEQ layers becomes very memory-efficient via the implicit function theorem. However, backpropagation through DEQ layers still requires solving an expensive Jacobian-based equation. In this paper, we introduce a simple but effective strategy to avoid this computational burden. Our method relies on the Jacobian approximation of Broyden's method after the forward pass to compute the gradients during the backward pass. Experiments show that simply re-using this approximation can significantly speed up the training while not causing any performance degradation.

Improving Self-Supervised Learning for Audio Representations by Feature Diversity and Decorrelation

Self-supervised learning (SSL) has recently shown remarkable results in closing the gap between supervised and unsupervised learning. The idea is to learn robust features that are invariant to distortions of the input data. Despite its success, this idea can suffer from a collapsing issue where the network produces a constant representation. To this end, we introduce SELFIE, a novel Self-supervised Learning approach for audio representation via Feature Diversity and Decorrelation. SELFIE avoids the collapsing issue by ensuring that the representation (i) maintains a high diversity among embeddings and (ii) decorrelates the dependencies between dimensions. SELFIE is pre-trained on the large-scale AudioSet dataset and its embeddings are validated on nine audio downstream tasks, including speech, music, and sound event recognition. Experimental results show that SELFIE outperforms existing SSL methods in several tasks.

Differentiable Duration Modeling for End-to-End Text-to-Speech

Parallel text-to-speech (TTS) models have recently enabled fast and highly-natural speech synthesis. However, such models typically require external alignment models, which are not necessarily optimized for the decoder as they are not jointly trained. In this paper, we propose a differentiable duration method for learning monotonic alignments between input and output sequences. Our method is based on a soft-duration mechanism that optimizes a stochastic process in expectation. Using this differentiable duration method, a direct text to waveform TTS model is introduced to produce raw audio as output instead of performing neural vocoding. Our model learns to perform high-fidelity speech synthesis through a combination of adversarial training and matching the total ground-truth duration. Experimental results show that our model obtains competitive results while enjoying a much simpler training pipeline. Audio samples are available online.

Neural Predictor for Black-Box Adversarial Attacks on Speech Recognition

Recent works have revealed the vulnerability of automatic speech recognition (ASR) models to adversarial examples (AEs), i.e., small perturbations that cause an error in the transcription of the audio signal. Studying audio adversarial attacks is therefore the first step towards robust ASR. Despite the significant progress made in attacking audio examples, the black-box attack remains challenging because only the hard-label information of transcriptions is provided. Due to this limited information, existing black-box methods often require an excessive number of queries to attack a single audio example. In this paper, we introduce NP-Attack, a neural predictor-based method, which progressively evolves the search towards a small adversarial perturbation. Given a perturbation direction, our neural predictor directly estimates the smallest perturbation that causes a mistranscription. In particular, it enables NP-Attack to accurately learn promising perturbation directions via gradient-based optimization. Experimental results show that NP-Attack achieves competitive results with other state-of-the-art black-box adversarial attacks while requiring a significantly smaller number of queries. The code of NP-Attack is available online.

NVC-Net: End-to-End Adversarial Voice Conversion

Voice conversion has gained increasing popularity in many applications of speech synthesis. The idea is to change the voice identity from one speaker into another while keeping the linguistic content unchanged. Many voice conversion approaches rely on the use of a vocoder to reconstruct the speech from acoustic features, and as a consequence, the speech quality heavily depends on such a vocoder. In this paper, we propose NVC-Net, an end-to-end adversarial network, which performs voice conversion directly on the raw audio waveform of arbitrary length. By disentangling the speaker identity from the speech content, NVC-Net is able to perform non-parallel traditional many-to-many voice conversion as well as zero-shot voice conversion from a short utterance of an unseen target speaker. Importantly, NVC-Net is non-autoregressive and fully convolutional, achieving fast inference. Our model is capable of producing samples at a rate of more than 3600 kHz on an NVIDIA V100 GPU, being orders of magnitude faster than state-of-the-art methods under the same hardware configurations. Objective and subjective evaluations on non-parallel many-to-many voice conversion tasks show that NVC-Net obtains competitive results with significantly fewer parameters.