United States

Fine-tuning pre-trained models has been ubiquitously proven to be effective in a wide range of NLP tasks. However, fine-tuning the whole model is parameter inefficient as it always yields an entirely new model for each task. Currently, many research works propose to only fine-tune a small portion of the parameters while keeping most of the parameters shared across different tasks. These methods achieve surprisingly good performance and are shown to be more stable than their corresponding fully fine-tuned counterparts. However, such kind of methods is still not well understood. Some natural questions arise: How does the parameter sparsity lead to promising performance? Why is the model more stable than the fully fine-tuned models? How to choose the tunable parameters? In this paper, we first categorize the existing methods into random approaches, rule-based approaches, and projection-based approaches based on how they choose which parameters to tune. Then, we show that all of the methods are actually sparse fine-tuned models and conduct a novel theoretical analysis of them. We indicate that the sparsity is actually imposing a regularization on the original model by controlling the upper bound of the stability. Such stability leads to better generalization capability which has been empirically observed in a lot of recent research works. On the other hand, the random and rule-based methods do not utilize task-specific data information while the projection-based approaches suffer from the projection discontinuity problem. To solve these problems, we propose a novel Second-order Approximation Method (SAM) which approximates the original problem with an analytically solvable optimization function. The tunable parameters are determined by directly optimizing the approximation function. We conduct extensive experiments on several tasks. The experimental results show that our proposed SAM model outperforms many strong baseline models and it also verifies our theoretical analysis.

AAAI 2023

On the Effectiveness of Parameter-Efficient Fine-Tuning

other foundations of speech & natural language processing

interpretability & analysis of nlp models

learning & optimization for snlp

other foundations of machine learning

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers. 
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities. This includes talks to attend in person, talks to attend remotely, breaks with colleagues and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

poster

Constructing accurate training tuples is crucial for unsupervised local descriptor learning, yet challenging due to the absence of patch labels. The state-of-the-art approach constructs tuples with heuristic rules, which struggle to precisely depict real-world patch transformations, in spite of enabling fast model convergence. A possible solution to alleviate the problem is the clustering-based approach, which can capture realistic patch variations and learn more accurate class decision boundaries, but suffers from slow model convergence. This paper presents HybridDesc, an unsupervised approach that learns powerful local descriptor models with fast convergence speed by combining the rule-based and clustering-based approaches to construct training tuples. In addition, HybridDesc also contributes two concrete enhancing mechanisms: (1) a Differentiable Hyperparameter Search (DHS) strategy to find the optimal hyperparameter setting of the rule-based approach so as to provide accurate prior for the clustering-based approach, (2) an On-Demand Clustering (ODC) method to reduce the clustering overhead of the clustering-based approach without eroding its advantage. Extensive experimental results show that HybridDesc can efficiently learn local descriptors that surpass existing unsupervised local descriptors and even rival competitive supervised ones.

Revisiting Unsupervised Local Descriptor Learning

Adding visible watermark into image is a common copyright protection method of medias. Meanwhile, public research on watermark removal can be utilized as an adversarial technology to help the further development of watermarking. 
Existing watermark removal methods mainly adopt multi-task learning networks, which locate the watermark and restore the background simultaneously. However, these approaches view the task as an image-to-image reconstruction problem, where they only impose supervision after the final output, making the high-level semantic features shared between different tasks. 
To this end, inspired by the two-stage coarse-refinement network, we propose a novel contrastive learning mechanism to disentangle the high-level embedding semantic information of the images and watermarks, driving the respective network branch more oriented.
Specifically, the proposed mechanism is leveraged for watermark image decomposition, which aims to decouple the clean image and watermark hints in the high-level embedding space. This can guarantee the learning representation of the restored image enjoy more task-specific cues.
In addition, we introduce a self-attention-based enhancement module, which promotes the network's ability to capture semantic information among different regions, leading to further improvement on the contrastive learning mechanism. 
To validate the effectiveness of our proposed method, extensive experiments are conducted on different challenging benchmarks. Experimental evaluations show that our approach can achieve state-of-the-art performance and yield high-quality images. The code is available at: https://github.com/lianchengmingjue/DENet. 

DENet: Disentangled Embedding Network for Visible Watermark Removal

Referring image segmentation segments an image from a language expression. With the aim of producing high-quality masks, existing methods often adopt iterative learning approaches that rely on RNNs or stacked attention layers to refine vision-language features. Despite their complexity, RNN-based methods are subject to specific encoder choices, while attention-based methods offer limited gains. In this work, we introduce a simple yet effective alternative for progressively learning discriminative multi-modal features. The core idea of our approach is to leverage a continuously updated query as the representation of the target object and at each iteration, strengthen multi-modal features strongly correlated to the query while weakening less related ones. As the query is initialized by language features and successively updated by object features, our algorithm gradually shifts from being localization-centric to being segmentation-centric. This strategy enables the incremental recovery of missing object parts and/or removal of extraneous parts throughout iterations. Compared to its counterparts, our method is more versatile and can be easily plugged into prior arts to consistently bring improvements. Experimental results on the challenging datasets of RefCOCO, RefCOCO+, and G-Ref demonstrate its advantage with respect to the state-of-the-art methods.

Semantics-Aware Dynamic Localization and Refinement for Referring Image Segmentation

We propose the Hierarchical Flow (HF) model constrained by isometric regularizations for manifold learning that combines manifold learning goals such as dimensionality reduction, inference, sampling, projection and density estimation into one unified framework. Our proposed HF model is regularized to not only produce embeddings preserving the geometric structure of the manifold, but also project samples onto the manifold in a manner conforming to the rigorous definition of projection. Theoretical guarantees are provided for our HF model to satisfy the two desired properties. In order to detect the real dimensionality of the manifold, we also propose a two-stage dimensionality reduction algorithm, which is a time-efficient algorithm thanks to the hierarchical architecture design of our HF model. Experimental results justify our theoretical analysis, demonstrate the superiority of our dimensionality reduction algorithm in cost of training time, and verify the effect of the aforementioned properties in improving performances on downstream tasks such as anomaly detection.

Isometric Manifold Learning using Hierarchical Flow

Cross-domain image retrieval aims at retrieving images across different domains to excavate cross-domain classificatory or correspondence relationships. This paper studies a less-touched problem of cross-domain image retrieval, \textit{i.e.}, unsupervised cross-domain image retrieval, considering the following practical assumptions: (i) no correspondence relationship, and (ii) no category annotations. It is challenging to align and bridge distinct domains without cross-domain correspondence. To tackle the challenge, we present a novel Correspondence-free Domain Alignment (\method) method to effectively eliminate the cross-domain gap through In-domain Self-matching Supervision (ISS) and Cross-domain Classifier Alignment (CCA). To be specific, ISS is presented to encapsulate discriminative information into the latent common space by elaborating a novel self-matching supervision mechanism. To alleviate the cross-domain discrepancy, CCA is proposed to align distinct domain-specific classifiers. Thanks to the ISS and CCA, our method could encode the discrimination into the domain-invariant embedding space for unsupervised cross-domain image retrieval. To verify the effectiveness of the proposed method, extensive experiments are conducted on four benchmark datasets compared with six state-of-the-art methods.

Correspondence-Free Domain Alignment for Unsupervised Cross-Domain Image Retrieval

Attention mechanism has become a standard fixture in many state-of-the-art natural language processing (NLP) models, not only due to its outstanding performance, but also because it provides plausible innate explanations for neural architectures. However, recent studies show that attention is unstable against randomness and perturbations during training or testing, such as random seeds and slight perturbation of embeddings, which impedes it from being a faithful explanation tool. Thus, a natural question is whether we can find an alternate of the vanilla attention, which is more stable and could keep the key characteristics of the explanation. In this paper, we provide a rigorous definition of such an attention method named SEAT (Stable and Explainable ATtention). Specifically, SEAT has the following three properties: (1) Its prediction distribution is close to the prediction of the vanilla attention; (2) Its top- k indices largely overlap with those of the vanilla attention; (3) It is robust w.r.t perturbations, i.e., any slight perturbation on SEAT will not change the attention and prediction distribution too much, which implicitly indicates that it is stable to randomness and perturbations. Furthermore, we propose an optimization method for obtaining SEAT, which could be considered as revising the vanilla attention. Finally, through in- tensive experiments on various datasets, we compare our SEAT with other baseline methods using RNN, BiLSTM and BERT architectures, with different evaluation metrics on model interpretation, stability and accuracy. Results show that, besides preserving the original explainability and model performance, SEAT is more stable against input perturbations and training randomness, which indicates it is a more faithful explanation.

SEAT: Stable and Explainable Attention

Instant runoff voting (IRV) is an increasingly-popular alternative to traditional plurality voting in which voters submit rankings over the candidates rather than single votes. In practice, elections using IRV often restrict the ballot length, the number of candidates a voter is allowed to rank on their ballot. We theoretically and empirically analyze how ballot length can influence the outcome of an election, given fixed voter preferences. We show that there exist preference profiles over $k$ candidates such that up to $k-1$ different candidates win at different ballot lengths. We derive exact lower bounds on the number of voters required for such profiles and provide a construction matching the lower bound for unrestricted voter preferences. Additionally, we characterize which sequences of winners are possible over ballot lengths and provide explicit profile constructions achieving any feasible winner sequence. We also examine how classic preference restrictions influence our results---for instance, single-peakedness makes $k-1$ different winners impossible but still allows at least $\Omega(\sqrt k)$. Finally, we analyze a collection of 168 real-world elections, where we truncate rankings to simulate shorter ballots. We find that shorter ballots could have changed the outcome in one quarter of these elections. Our results highlight ballot length as a consequential degree of freedom in the design of IRV elections.


Ballot Length in Instant Runoff Voting

Multi-player multi-armed bandit is an increasingly relevant decision-making problem, motivated by applications to cognitive radio systems.  Most research for this problem focuses exclusively on the settings that players have full access to all arms and receive no reward when pulling the same arm.  Hence all players solve the same bandit problem with the goal of maximizing their cumulative reward. However, these settings neglect several important factors in many real-world applications, where players have limited access to a dynamic local subset of arms (i.e., an arm could sometimes be ``walking'' and not accessible to the player).  To this end, this paper proposes a multi-player multi-armed walking bandits model, aiming to address aforementioned modeling issues. The goal now is to maximize the reward, however, players can only pull arms from the local subset and only collect a full reward if no other players pull the same arm.  We adopt Upper Confidence Bound (UCB) to deal with the exploration-exploitation tradeoff and employ distributed optimization techniques to properly handle collisions.  By carefully integrating these two techniques, we propose a decentralized algorithm with near-optimal guarantee on the regret, and can be easily implemented to obtain competitive empirical performance.

Decentralized Stochastic Multi-Player Multi-Armed Walking Bandits

Mathematical reasoning is one of the crucial abilities of general artificial intelligence, which requires machines to master mathematical logic and knowledge from solving problems. However, existing approaches are not transparent (thus not interpretable) in terms of what knowledge has been learned and applied in the reasoning process. In this paper, we propose a general Learning by Applying (LeAp) framework to enhance existing models (backbones) in a principled way by explicit knowledge learning. In LeAp, we perform knowledge learning in a novel problem-knowledge-expression paradigm, with a Knowledge Encoder to acquire knowledge from problem data and a Knowledge Decoder to apply knowledge for expression reasoning. The learned mathematical knowledge, including word-word relations and word-operator relations, forms an explicit knowledge graph, which bridges the knowledge “learning” and “applying” organically. Moreover, for problem solving, we design a semantics-enhanced module and a reasoning-enhanced module that apply knowledge to improve the problem comprehension and symbol reasoning abilities of any backbone, respectively. We theoretically prove the superiority of LeAp's autonomous learning mechanism. Experiments on three real-world datasets show that LeAp improves all backbones' performances, learns accurate knowledge, and achieves a more interpretable reasoning process.

Learning by Applying: A General Framework for Mathematical Reasoning via Enhancing Explicit Knowledge Learning

In image classification, debiasing aims to train a classifier to be less susceptible to dataset bias, the strong correlation between peripheral attributes of data samples and a target class. For example, even if the frog class in the dataset mainly consists of frog images with a swamp background (i.e., bias-aligned samples), a debiased classifier should be able to correctly classify a frog at a beach (i.e., bias-conflicting samples). Recent debiasing approaches commonly use two components for debiasing, a biased model $f_B$ and a debiased model $f_D$. $f_B$ is trained to focus on bias-aligned samples (i.e., overfitted to the bias) while $f_D$ is mainly trained with bias-conflicting samples by concentrating on samples which $f_B$ fails to learn, leading $f_D$ to be less susceptible to the dataset bias. While the state-of-the-art debiasing techniques have aimed to better train $f_D$, we focus on training $f_B$, an overlooked component until now. Our empirical analysis reveals that removing the bias-conflicting samples from the training set for $f_B$ is important for improving the debiasing performance of $f_D$. This is due to the fact that the bias-conflicting samples work as noisy samples for amplifying the bias for $f_B$ since those samples do not include the bias attribute. To this end, we propose a simple yet effective data sample selection method which removes the bias-conflicting samples to construct a bias-amplified dataset for training $f_B$. Our data sample selection method can be directly applied to existing reweighting-based debiasing approaches, obtaining consistent performance boost and achieving the state-of-the-art performance on both synthetic and real-world datasets. 

Premium content

On the Effectiveness of Parameter-Efficient Fine-Tuning

Downloads

Next from AAAI 2023

Revisiting Unsupervised Local Descriptor Learning

Similar lecture

Don't Be So Sure! Boosting ASR Decoding via Confidence Relaxation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES