Nicholas Apostoloff

Xavier Suau, Pieter Delobelle, Katherine Metcalf et al.

An important issue with Large Language Models (LLMs) is their undesired ability to generate toxic language. In this work, we show that the neurons responsible for toxicity can be determined by their power to discriminate toxic sentences, and that toxic language can be mitigated by reducing their activation levels proportionally to this power. We propose AUROC adaptation (AurA), an intervention that can be applied to any pre-trained LLM to mitigate toxicity. As the intervention is proportional to the ability of each neuron to discriminate toxic content, it is free of any model-dependent hyperparameters. We show that AurA can achieve up to $2.2 \times$ reduction in toxicity with only a $0.72$ perplexity increase. We also show that AurA is effective with models of different scale (from 1.5B to 40B parameters), and its effectiveness in mitigating toxic language, while preserving common-sense zero-shot abilities, holds across all scales. AurA can be combined with pre-prompting strategies, boosting its average mitigation potential from $1.28\times$ to $2.35\times$. Moreover, AurA can counteract adversarial pre-prompts that maliciously elicit toxic content, making it an effective method for deploying safer and less toxic models.

Zakaria Aldeneh, Masha Fedzechkina, Skyler Seto et al. · apple-ml

Generating realistic lip motion from audio to simulate speech production is critical for driving natural character animation. Previous research has shown that traditional metrics used to optimize and assess models for generating lip motion from speech are not a good indicator of subjective opinion of animation quality. Devising metrics that align with subjective opinion first requires understanding what impacts human perception of quality. In this work, we focus on the degree of articulation and run a series of experiments to study how articulation strength impacts human perception of lip motion accompanying speech. Specifically, we study how increasing under-articulated (dampened) and over-articulated (exaggerated) lip motion affects human perception of quality. We examine the impact of articulation strength on human perception when considering only lip motion, where viewers are presented with talking faces represented by landmarks, and in the context of embodied characters, where viewers are presented with photo-realistic videos. Our results show that viewers prefer over-articulated lip motion consistently more than under-articulated lip motion and that this preference generalizes across different speakers and embodiments.

Miguel Sarabia, Elena Menyaylenko, Alessandro Toso et al.

We present Spatial LibriSpeech, a spatial audio dataset with over 650 hours of 19-channel audio, first-order ambisonics, and optional distractor noise. Spatial LibriSpeech is designed for machine learning model training, and it includes labels for source position, speaking direction, room acoustics and geometry. Spatial LibriSpeech is generated by augmenting LibriSpeech samples with 200k+ simulated acoustic conditions across 8k+ synthetic rooms. To demonstrate the utility of our dataset, we train models on four spatial audio tasks, resulting in a median absolute error of 6.60° on 3D source localization, 0.43m on distance, 90.66ms on T30, and 2.74dB on DRR estimation. We show that the same models generalize well to widely-used evaluation datasets, e.g., obtaining a median absolute error of 12.43° on 3D source localization on TUT Sound Events 2018, and 157.32ms on T30 estimation on ACE Challenge.

Yinong Oliver Wang, Nivedha Sivakumar, Falaah Arif Khan et al.

The recent rapid adoption of large language models (LLMs) highlights the critical need for benchmarking their fairness. Conventional fairness metrics, which focus on discrete accuracy-based evaluations (i.e., prediction correctness), fail to capture the implicit impact of model uncertainty (e.g., higher model confidence about one group over another despite similar accuracy). To address this limitation, we propose an uncertainty-aware fairness metric, UCerF, to enable a fine-grained evaluation of model fairness that is more reflective of the internal bias in model decisions compared to conventional fairness measures. Furthermore, observing data size, diversity, and clarity issues in current datasets, we introduce a new gender-occupation fairness evaluation dataset with 31,756 samples for co-reference resolution, offering a more diverse and suitable dataset for evaluating modern LLMs. We establish a benchmark, using our metric and dataset, and apply it to evaluate the behavior of ten open-source LLMs. For example, Mistral-7B exhibits suboptimal fairness due to high confidence in incorrect predictions, a detail overlooked by Equalized Odds but captured by UCerF. Overall, our proposed LLM benchmark, which evaluates fairness with uncertainty awareness, paves the way for developing more transparent and accountable AI systems.

Pau Rodriguez, Arno Blaas, Michal Klein et al.

The increasing capabilities of large generative models and their ever more widespread deployment have raised concerns about their reliability, safety, and potential misuse. To address these issues, recent works have proposed to control model generation by steering model activations in order to effectively induce or prevent the emergence of concepts or behaviors in the generated output. In this paper we introduce Activation Transport (AcT), a general framework to steer activations guided by optimal transport theory that generalizes many previous activation-steering works. AcT is modality-agnostic and provides fine-grained control over the model behavior with negligible computational overhead, while minimally impacting model abilities. We experimentally show the effectiveness and versatility of our approach by addressing key challenges in large language models (LLMs) and text-to-image diffusion models (T2Is). For LLMs, we show that AcT can effectively mitigate toxicity, induce arbitrary concepts, and increase their truthfulness. In T2Is, we show how AcT enables fine-grained style control and concept negation.

Natalie Mackraz, Nivedha Sivakumar, Samira Khorshidi et al.

Large language models (LLMs) are increasingly being adapted to achieve task-specificity for deployment in real-world decision systems. Several previous works have investigated the bias transfer hypothesis (BTH) by studying the effect of the fine-tuning adaptation strategy on model fairness to find that fairness in pre-trained masked language models have limited effect on the fairness of models when adapted using fine-tuning. In this work, we expand the study of BTH to causal models under prompt adaptations, as prompting is an accessible, and compute-efficient way to deploy models in real-world systems. In contrast to previous works, we establish that intrinsic biases in pre-trained Mistral, Falcon and Llama models are strongly correlated (rho >= 0.94) with biases when the same models are zero- and few-shot prompted, using a pronoun co-reference resolution task. Further, we find that bias transfer remains strongly correlated even when LLMs are specifically prompted to exhibit fair or biased behavior (rho >= 0.92), and few-shot length and stereotypical composition are varied (rho >= 0.97). Our findings highlight the importance of ensuring fairness in pre-trained LLMs, especially when they are later used to perform downstream tasks via prompt adaptation.

Nivedha Sivakumar, Natalie Mackraz, Samira Khorshidi et al.

A dangerous assumption that can be made from prior work on the bias transfer hypothesis (BTH) is that biases do not transfer from pre-trained large language models (LLMs) to adapted models. We invalidate this assumption by studying the BTH in causal models under prompt adaptations, as prompting is an extremely popular and accessible adaptation strategy used in real-world applications. In contrast to prior work, we find that biases can transfer through prompting and that popular prompt-based mitigation methods do not consistently prevent biases from transferring. Specifically, the correlation between intrinsic biases and those after prompt adaptation remain moderate to strong across demographics and tasks -- for example, gender (rho >= 0.94) in co-reference resolution, and age (rho >= 0.98) and religion (rho >= 0.69) in question answering. Further, we find that biases remain strongly correlated when varying few-shot composition parameters, such as sample size, stereotypical content, occupational distribution and representational balance (rho >= 0.90). We evaluate several prompt-based debiasing strategies and find that different approaches have distinct strengths, but none consistently reduce bias transfer across models, tasks or demographics. These results demonstrate that correcting bias, and potentially improving reasoning ability, in intrinsic models may prevent propagation of biases to downstream tasks.

Lucas Monteiro Paes, Nivedha Sivakumar, Yinong Oliver Wang et al.

Generative models are often deployed to make decisions on behalf of users, such as vision-language models (VLMs) identifying which person in a room is a doctor to help visually impaired individuals. Yet, VLM decisions are influenced by the perceived demographic attributes of people in the input, which can lead to biased outcomes like failing to identify women as doctors. Moreover, when reducing bias leads to performance loss, users may have varying needs for balancing bias mitigation with overall model capabilities, highlighting the demand for methods that enable controllable bias reduction during inference. Activation steering is a popular approach for inference-time controllability that has shown potential in inducing safer behavior in large language models (LLMs). However, we observe that current steering methods struggle to correct biases, where equiprobable outcomes across demographic groups are required. To address this, we propose Direct Steering Optimization (DSO) which uses reinforcement learning to find linear transformations for steering activations, tailored to mitigate bias while maintaining control over model performance. We demonstrate that DSO achieves state-of-the-art trade-off between fairness and capabilities on both VLMs and LLMs, while offering practitioners inference-time control over the trade-off. Overall, our work highlights the benefit of designing steering strategies that are directly optimized to control model behavior, providing more effective bias intervention than methods that rely on pre-defined heuristics for controllability.

Falaah Arif Khan, Nivedha Sivakumar, Yinong Oliver Wang et al.

Large language models (LLMs) have achieved impressive performance, leading to their widespread adoption as decision-support tools in resource-constrained contexts like hiring and admissions. There is, however, scientific consensus that AI systems can reflect and exacerbate societal biases, raising concerns about identity-based harm when used in critical social contexts. Prior work has laid a solid foundation for assessing bias in LLMs by evaluating demographic disparities in different language reasoning tasks. In this work, we extend single-axis fairness evaluations to examine intersectional bias, recognizing that when multiple axes of discrimination intersect, they create distinct patterns of disadvantage. We create a new benchmark called WinoIdentity by augmenting the WinoBias dataset with 25 demographic markers across 10 attributes, including age, nationality, and race, intersected with binary gender, yielding 245,700 prompts to evaluate 50 distinct bias patterns. Focusing on harms of omission due to underrepresentation, we investigate bias through the lens of uncertainty and propose a group (un)fairness metric called Coreference Confidence Disparity which measures whether models are more or less confident for some intersectional identities than others. We evaluate five recently published LLMs and find confidence disparities as high as 40% along various demographic attributes including body type, sexual orientation and socio-economic status, with models being most uncertain about doubly-disadvantaged identities in anti-stereotypical settings. Surprisingly, coreference confidence decreases even for hegemonic or privileged markers, indicating that the recent impressive performance of LLMs is more likely due to memorization than logical reasoning. Notably, these are two independent failures in value alignment and validity that can compound to cause social harm.

Krishna Patel, Nivedha Sivakumar, Barry-John Theobald et al.

We investigate fairness dynamics during Large Language Model (LLM) training to enable the diagnoses of biases and mitigations through training interventions like early stopping; we find that biases can emerge suddenly and do not always follow common performance metrics. We introduce two new metrics to evaluate fairness dynamics holistically during model pre-training: Average Rank and Jensen-Shannon Divergence by Parts. These metrics provide insights into the Pythia models' progression of biases in gender prediction of occupations on the WinoBias dataset. By monitoring these dynamics, we find that (1) Pythia-6.9b is biased towards men; it becomes more performant and confident predicting "male" than "female" during training, (2) via early-stopping, Pythia-6.9b can exchange 1.7% accuracy on LAMBADA for a 92.5% increase in fairness, and (3) larger models can exhibit more bias; Pythia-6.9b makes more assumptions about gender than Pythia-160m, even when a subject's gender is not specified.

Andrew Silva, Katherine Metcalf, Nicholas Apostoloff et al.

Federated learning enables the deployment of machine learning to problems for which centralized data collection is impractical. Adding differential privacy guarantees bounds on privacy while data are contributed to a global model. Adding personalization to federated learning introduces new challenges as we must account for preferences of individual users, where a data sample could have conflicting labels because one sub-population of users might view an input positively, but other sub-populations view the same input negatively. We present FedEmbed, a new approach to private federated learning for personalizing a global model that uses (1) sub-populations of similar users, and (2) personal embeddings. We demonstrate that current approaches to federated learning are inadequate for handling data with conflicting labels, and we show that FedEmbed achieves up to 45% improvement over baseline approaches to personalized private federated learning.

Aparna R. Joshi, Xavier Suau, Nivedha Sivakumar et al.

As the use of deep learning in high impact domains becomes ubiquitous, it is increasingly important to assess the resilience of models. One such high impact domain is that of face recognition, with real world applications involving images affected by various degradations, such as motion blur or high exposure. Moreover, images captured across different attributes, such as gender and race, can also challenge the robustness of a face recognition algorithm. While traditional summary statistics suggest that the aggregate performance of face recognition models has continued to improve, these metrics do not directly measure the robustness or fairness of the models. Visual Psychophysics Sensitivity Analysis (VPSA) [1] provides a way to pinpoint the individual causes of failure by way of introducing incremental perturbations in the data. However, perturbations may affect subgroups differently. In this paper, we propose a new fairness evaluation based on robustness in the form of a generic framework that extends VPSA. With this framework, we can analyze the ability of a model to perform fairly for different subgroups of a population affected by perturbations, and pinpoint the exact failure modes for a subgroup by measuring targeted robustness. With the increasing focus on the fairness of models, we use face recognition as an example application of our framework and propose to compactly visualize the fairness analysis of a model via AUC matrices. We analyze the performance of common face recognition models and empirically show that certain subgroups are at a disadvantage when images are perturbed, thereby uncovering trends that were not visible using the model's performance on subgroups without perturbations.

Bobby Yan, Skyler Seto, Nicholas Apostoloff

Machine learning models are trained to minimize the mean loss for a single metric, and thus typically do not consider fairness and robustness. Neglecting such metrics in training can make these models prone to fairness violations when training data are imbalanced or test distributions differ. This work introduces Fairness Optimized Reweighting via Meta-Learning (FORML), a training algorithm that balances fairness and robustness with accuracy by jointly learning training sample weights and neural network parameters. The approach increases model fairness by learning to balance the contributions from both over- and under-represented sub-groups through dynamic reweighting of the data learned from a user-specified held-out set representative of the distribution under which fairness is desired. FORML improves equality of opportunity fairness criteria on image classification tasks, reduces bias of corrupted labels, and facilitates building more fair datasets via data condensation. These improvements are achieved without pre-processing data or post-processing model outputs, without learning an additional weighting function, without changing model architecture, and while maintaining accuracy on the original predictive metric.

Arno Blaas, Xavier Suau, Jason Ramapuram et al.

Image augmentations applied during training are crucial for the generalization performance of image classifiers. Therefore, a large body of research has focused on finding the optimal augmentation policy for a given task. Yet, RandAugment [2], a simple random augmentation policy, has recently been shown to outperform existing sophisticated policies. Only Adversarial AutoAugment (AdvAA) [11], an approach based on the idea of adversarial training, has shown to be better than RandAugment. In this paper, we show that random augmentations are still competitive compared to an optimal adversarial approach, as well as to simple curricula, and conjecture that the success of AdvAA is due to the stochasticity of the policy controller network, which introduces a mild form of curriculum.

Xavier Suau, Luca Zappella, Nicholas Apostoloff

In this paper we aim to investigate the mechanisms that guide text generation with pre-trained Transformer-based Language Models (TLMs). Grounded on the Product of Experts formulation by Hinton (1999), we describe a generative mechanism that exploits expert units which naturally exist in TLMs. Such units are responsible for detecting concepts in the input and conditioning text generation on such concepts. We describe how to identify expert units and how to activate them during inference in order to induce any desired concept in the generated output. We find that the activation of a surprisingly small amount of units is sufficient to steer text generation (as little as 3 units in a model with 345M parameters). While the objective of this work is to learn more about how TLMs work, we show that our method is effective for conditioning without fine-tuning or using extra parameters, even on fine-grained homograph concepts. Additionally, we show that our method can be used to correct gender bias present in the output of TLMs and achieves gender parity for all evaluated contexts. We compare our method with FUDGE and PPLM-BoW, and show that our approach is able to achieve gender parity at a lower perplexity. The proposed method is accessible to a wide audience thanks to its simplicity and minimal compute needs. The findings in this paper are a step forward in understanding the generative mechanisms of TLMs.

Andrew Silva, Barry-John Theobald, Nicholas Apostoloff

Automatic speech recognition (ASR) is widely used in consumer electronics. ASR greatly improves the utility and accessibility of technology, but usually the output is only word sequences without punctuation. This can result in ambiguity in inferring user-intent. We first present a transformer-based approach for punctuation prediction that achieves 8% improvement on the IWSLT 2012 TED Task, beating the previous state of the art [1]. We next describe our multimodal model that learns from both text and audio, which achieves 8% improvement over the text-only algorithm on an internal dataset for which we have both the audio and transcriptions. Finally, we present an approach to learning a model using contextual dropout that allows us to handle variable amounts of future context at test time.

Nataniel Ruiz, Barry-John Theobald, Anurag Ranjan et al.

To detect bias in face recognition networks, it can be useful to probe a network under test using samples in which only specific attributes vary in some controlled way. However, capturing a sufficiently large dataset with specific control over the attributes of interest is difficult. In this work, we describe a simulator that applies specific head pose and facial expression adjustments to images of previously unseen people. The simulator first fits a 3D morphable model to a provided image, applies the desired head pose and facial expression controls, then renders the model into an image. Next, a conditional Generative Adversarial Network (GAN) conditioned on the original image and the rendered morphable model is used to produce the image of the original person with the new facial expression and head pose. We call this conditional GAN -- MorphGAN. Images generated using MorphGAN conserve the identity of the person in the original image, and the provided control over head pose and facial expression allows test sets to be created to identify robustness issues of a facial recognition deep network with respect to pose and expression. Images generated by MorphGAN can also serve as data augmentation when training data are scarce. We show that by augmenting small datasets of faces with new poses and expressions improves the recognition performance by up to 9% depending on the augmentation and data scarcity.

Ahmed Hussen Abdelaziz, Barry-John Theobald, Paul Dixon et al.

We describe our novel deep learning approach for driving animated faces using both acoustic and visual information. In particular, speech-related facial movements are generated using audiovisual information, and non-speech facial movements are generated using only visual information. To ensure that our model exploits both modalities during training, batches are generated that contain audio-only, video-only, and audiovisual input features. The probability of dropping a modality allows control over the degree to which the model exploits audio and visual information during training. Our trained model runs in real-time on resource limited hardware (e.g.\ a smart phone), it is user agnostic, and it is not dependent on a potentially error-prone transcription of the speech. We use subjective testing to demonstrate: 1) the improvement of audiovisual-driven animation over the equivalent video-only approach, and 2) the improvement in the animation of speech-related facial movements after introducing modality dropout. Before introducing dropout, viewers prefer audiovisual-driven animation in 51% of the test sequences compared with only 18% for video-driven. After introducing dropout viewer preference for audiovisual-driven animation increases to 74%, but decreases to 8% for video-only.

Xavier Suau, Luca Zappella, Nicholas Apostoloff

In this work we study the presence of expert units in pre-trained Transformer Models (TM), and how they impact a model's performance. We define expert units to be neurons that are able to classify a concept with a given average precision, where a concept is represented by a binary set of sentences containing the concept (or not). Leveraging the OneSec dataset (Scarlini et al., 2019), we compile a dataset of 1641 concepts that allows diverse expert units in TM to be discovered. We show that expert units are important in several ways: (1) The presence of expert units is correlated ($r^2=0.833$) with the generalization power of TM, which allows ranking TM without requiring fine-tuning on suites of downstream tasks. We further propose an empirical method to decide how accurate such experts should be to evaluate generalization. (2) The overlap of top experts between concepts provides a sensible way to quantify concept co-learning, which can be used for explainability of unknown concepts. (3) We show how to self-condition off-the-shelf pre-trained language models to generate text with a given concept by forcing the top experts to be active, without requiring re-training the model or using additional parameters.

Ahmed Hussen Abdelaziz, Barry-John Theobald, Justin Binder et al.

Speech-driven visual speech synthesis involves mapping features extracted from acoustic speech to the corresponding lip animation controls for a face model. This mapping can take many forms, but a powerful approach is to use deep neural networks (DNNs). However, a limitation is the lack of synchronized audio, video, and depth data required to reliably train the DNNs, especially for speaker-independent models. In this paper, we investigate adapting an automatic speech recognition (ASR) acoustic model (AM) for the visual speech synthesis problem. We train the AM on ten thousand hours of audio-only data. The AM is then adapted to the visual speech synthesis domain using ninety hours of synchronized audio-visual speech. Using a subjective assessment test, we compared the performance of the AM-initialized DNN to one with a random initialization. The results show that viewers significantly prefer animations generated from the AM-initialized DNN than the ones generated using the randomly initialized model. We conclude that visual speech synthesis can significantly benefit from the powerful representation of speech in the ASR acoustic models.

Katherine Metcalf, Barry-John Theobald, Garrett Weinberg et al.

We describe experiments towards building a conversational digital assistant that considers the preferred conversational style of the user. In particular, these experiments are designed to measure whether users prefer and trust an assistant whose conversational style matches their own. To this end we conducted a user study where subjects interacted with a digital assistant that responded in a way that either matched their conversational style, or did not. Using self-reported personality attributes and subjects' feedback on the interactions, we built models that can reliably predict a user's preferred conversational style.

Katherine Metcalf, Barry-John Theobald, Nicholas Apostoloff

We propose a method for modeling and learning turn-taking behaviors for accessing a shared resource. We model the individual behavior for each agent in an interaction and then use a multi-agent fusion model to generate a summary over the expected actions of the group to render the model independent of the number of agents. The individual behavior models are weighted finite state transducers (WFSTs) with weights dynamically updated during interactions, and the multi-agent fusion model is a logistic regression classifier. We test our models in a multi-agent tower-building environment, where a Q-learning agent learns to interact with rule-based agents. Our approach accurately models the underlying behavior patterns of the rule-based agents with accuracy ranging between 0.63 and 1.0 depending on the stochasticity of the other agent behaviors. In addition we show using KL-divergence that the model accurately captures the distribution of next actions when interacting with both a single agent (KL-divergence < 0.1) and with multiple agents (KL-divergence < 0.37). Finally, we demonstrate that our behavior model can be used by a Q-learning agent to take turns in an interactive turn-taking environment.

Xavier Suau, Luca Zappella, Nicholas Apostoloff

In this paper we introduce Principal Filter Analysis (PFA), an easy to use and effective method for neural network compression. PFA exploits the correlation between filter responses within network layers to recommend a smaller network that maintain as much as possible the accuracy of the full model. We propose two algorithms: the first allows users to target compression to specific network property, such as number of trainable variable (footprint), and produces a compressed model that satisfies the requested property while preserving the maximum amount of spectral energy in the responses of each layer, while the second is a parameter-free heuristic that selects the compression used at each layer by trying to mimic an ideal set of uncorrelated responses. Since PFA compresses networks based on the correlation of their responses we show in our experiments that it gains the additional flexibility of adapting each architecture to a specific domain while compressing. PFA is evaluated against several architectures and datasets, and shows considerable compression rates without compromising accuracy, e.g., for VGG-16 on CIFAR-10, CIFAR-100 and ImageNet, PFA achieves a compression rate of 8x, 3x, and 1.4x with an accuracy gain of 0.4%, 1.4% points, and 2.4% respectively. Our tests show that PFA is competitive with state-of-the-art approaches while removing adoption barriers thanks to its practical implementation, intuitive philosophy and ease of use.