Publications by Angelo Porrello

The adoption of Multi-Instance Learning (MIL) for classifying Whole-Slide Images (WSIs) has increased in recent years. Indeed, pixel-level annotation of gigapixel WSI is mostly unfeasible and time-consuming in practice. For this reason, MIL approaches have been profitably integrated with the most recent deep-learning solutions for WSI classification to support clinical practice and diagnosis. Nevertheless, the majority of such approaches overlook the multi-scale nature of the WSIs; the few existing hierarchical MIL proposals simply flatten the multi-scale representations by concatenation or summation of features vectors, neglecting the spatial structure of the WSI. Our work aims to unleash the full potential of pyramidal structured WSI; to do so, we propose a graph-based multi-scale MIL approach, termed DAS-MIL, that exploits message passing to let information flows across multiple scales. By means of a knowledge distillation schema, the alignment between the latent space representation at different resolutions is encouraged while preserving the diversity in the informative content. The effectiveness of the proposed framework is demonstrated on two well-known datasets, where we outperform SOTA on WSI classification, gaining a +1.9% AUC and +3.3¬curacy on the popular Camelyon16 benchmark.

Denoising Diffusion Probabilistic Models have shown an impressive generation quality although their long sampling chain leads to high computational costs. In this paper, we observe that a long sampling chain also leads to an error accumulation phenomenon, which is similar to the exposure bias problem in autoregressive text generation. Specifically, we note that there is a discrepancy between training and testing, since the former is conditioned on the ground truth samples, while the latter is conditioned on the previously generated results. To alleviate this problem, we propose a very simple but effective training regularization, consisting in perturbing the ground truth samples to simulate the inference time prediction errors. We empirically show that, without affecting the recall and precision, the proposed input perturbation leads to a significant improvement in the sample quality while reducing both the training and the inference times. For instance, on CelebA 64×64, we achieve a new state-of-the-art FID score of 1.27, while saving 37.5% of the training time. The code is available at https://github.com/forever208/DDPM-IP.

Artificial-intelligence-based methods are regularly used in the biomedical sciences, mainly in the field of diagnostic imaging. Recently, convolutional neural networks have been trained to score pleurisy and pneumonia in slaughtered pigs. The aim of this study is to further evaluate the performance of a convolutional neural network when compared with the gold standard (i.e., scores provided by a skilled operator along the slaughter chain through visual inspection and palpation). In total, 441 lungs (180 healthy and 261 diseased) are included in this study. Each lung was scored according to traditional methods, which represent the gold standard (Madec’s and Christensen’s grids). Moreover, the same lungs were photographed and thereafter scored by a trained convolutional neural network. Overall, the results reveal that the convolutional neural network is very specific (95.55%) and quite sensitive (85.05%), showing a rather high correlation when compared with the scores provided by a skilled veterinarian (Spearman’s coefficient = 0.831, p < 0.01). In summary, this study suggests that convolutional neural networks could be effectively used at slaughterhouses and stimulates further investigation in this field of research.

The field of multi-object tracking has recently seen a renewed interest in the good old schema of tracking-by-detection, as its simplicity and strong priors spare it from the complex design and painful babysitting of tracking-by-attention approaches. In view of this, we aim at extending tracking-by-detection to multi-modal settings, where a comprehensive cost has to be computed from heterogeneous information e.g., 2D motion cues, visual appearance, and pose estimates. More precisely, we follow a case study where a rough estimate of 3D information is also available and must be merged with other traditional metrics (e.g., the IoU). To achieve that, recent approaches resort to either simple rules or complex heuristics to balance the contribution of each cost. However, i) they require careful tuning of tailored hyperparameters on a hold-out set, and ii) they imply these costs to be independent, which does not hold in reality. We address these issues by building upon an elegant probabilistic formulation, which considers the cost of a candidate association as the negative log-likelihood yielded by a deep density estimator, trained to model the conditional joint probability distribution of correct associations. Our experiments, conducted on both simulated and real benchmarks, show that our approach consistently enhances the performance of several tracking-by-detection algorithms.

Continual Learning (CL) investigates how to train Deep Networks on a stream of tasks without incurring forgetting. CL settings proposed in literature assume that every incoming example is paired with ground-truth annotations. However, this clashes with many real-world applications: gathering labeled data, which is in itself tedious and expensive, becomes infeasible when data flow as a stream. This work explores Continual Semi-Supervised Learning (CSSL): here, only a small fraction of labeled input examples are shown to the learner. We assess how current CL methods (e.g.: EWC, LwF, iCaRL, ER, GDumb, DER) perform in this novel and challenging scenario, where overfitting entangles forgetting. Subsequently, we design a novel CSSL method that exploits metric learning and consistency regularization to leverage unlabeled examples while learning. We show that our proposal exhibits higher resilience to diminishing supervision and, even more surprisingly, relying only on supervision suffices to outperform SOTA methods trained under full supervision.

Accurate prediction of future human positions is an essential task for modern video-surveillance systems. Current state-of-the-art models usually rely on a "history" of past tracked locations (e.g., 3 to 5 seconds) to predict a plausible sequence of future locations (e.g., up to the next 5 seconds). We feel that this common schema neglects critical traits of realistic applications: as the collection of input trajectories involves machine perception (i.e., detection and tracking), incorrect detection and fragmentation errors may accumulate in crowded scenes, leading to tracking drifts. On this account, the model would be fed with corrupted and noisy input data, thus fatally affecting its prediction performance.In this regard, we focus on delivering accurate predictions when only few input observations are used, thus potentially lowering the risks associated with automatic perception. To this end, we conceive a novel distillation strategy that allows a knowledge transfer from a teacher network to a student one, the latter fed with fewer observations (just two ones). We show that a properly defined teacher supervision allows a student network to perform comparably to state-of-the-art approaches that demand more observations. Besides, extensive experiments on common trajectory forecasting datasets highlight that our student network better generalizes to unseen scenarios.

Rehearsal approaches enjoy immense popularity with Continual Learning (CL) practitioners. These methods collect samples from previously encountered data distributions in a small memory buffer; subsequently, they repeatedly optimize on the latter to prevent catastrophic forgetting. This work draws attention to a hidden pitfall of this widespread practice: repeated optimization on a small pool of data inevitably leads to tight and unstable decision boundaries, which are a major hindrance to generalization. To address this issue, we propose Lipschitz-DrivEn Rehearsal (LiDER), a surrogate objective that induces smoothness in the backbone network by constraining its layer-wise Lipschitz constants w.r.t. replay examples. By means of extensive experiments, we show that applying LiDER delivers a stable performance gain to several state-of-the-art rehearsal CL methods across multiple datasets, both in the presence and absence of pre-training. Through additional ablative experiments, we highlight peculiar aspects of buffer overfitting in CL and better characterize the effect produced by LiDER. Code is available at https://github.com/aimagelab/LiDER

Nell'ultimo decennio, il Deep Learning è diventato un argomento caldo oltre che uno strumento dirompente nel contesto del Machine Learning e della Computer Vision. Si basa su un paradigma di apprendimento in cui i dati (ad esempio, i video acquisiti da telecamere di video-sorveglianza poste su una strada pubblica) giocano un ruolo cruciale. Sfruttando un gran numero di esempi, è possibile imparare compiti complessi e simili a quelli svolti da esseri umani (ad esempio, riconoscere azioni anomale in un video-stream) con risultati impressionanti. Tuttavia, se la disponibilità di dati rappresenta la più grande forza delle tecniche di Deep Learning, essa nasconde anche la più grande debolezza: lo sviluppo di applicazioni e servizi è, infatti, spesso limitato da tale requisito, poiché l'acquisizione e il mantenimento di una enorme quantità di dati sono attività costose che richiedono personale esperto e attrezzature idonee. Tuttavia, la progettazione delle moderne architetture di Deep Learning offre diversi gradi di libertà, i quali possono essere sfruttati per mitigare la mancanza di dati di allenamento, sia essa parziale che completa. L'idea di fondo è quella di compensare tale mancanza incorporando una conoscenza preliminare che gli umani (in particolare, colore che controllano e guidano il processo di apprendimento) detengono sul dominio in questione. Infatti, le regole e le proprietà intrinseche si estendono ben oltre i dati di formazione e spesso possono essere identificate e imposte al modello di learning. Se prendiamo in considerazione la classificazione delle immagini, il successo delle Reti Neurali Convoluzionali (CNN) rispetto alle soluzioni del passato (come le Reti Neurali Multistrato) può essere attribuito principalmente a tale pratica. Infatti, i principi di progettazione del suo elemento costitutivo fondamentale (cioè la convoluzione tra due segnali 2D) riflettono naturalmente ciò che sapevamo sulle immagini naturali: le correlazioni che sussistono tra le regioni vicine dell'immagine hanno fornito pertanto una potente intuizione per lo sviluppo di modelli efficienti ed efficaci come lo sono ancora le CNN. Lo scopo di questa tesi riguarda l'indagine e la proposta di nuovi modi di modellare e iniettare la conoscenza a priori nelle architetture di Deep Learning. È importante sottolineare che tale discussione è trasversale: infatti, si concentra su diversi domini di dati (ad esempio, immagini, video, dati strutturati mediante un grafo, ecc.) e coinvolge diversi livelli della pipeline complessiva. Su quest'ultimo punto, il lettore viene guidato in questa ricerca attraverso la seguente triplice categorizzazione: i) approcci basati sui parametri, che limitano lo spazio delle soluzioni possibili a quelle regioni che riflettono le proprietà geometriche dei dati; ii) approcci goal-driven, che guidano il processo di apprendimento verso soluzioni che incarnano alcune proprietà vantaggiose; iii) approcci data-driven, che sfruttano i dati per estrarre la conoscenza da utilizzare successivamente per condizionare l'algoritmo di training. Insieme a una descrizione completa di entrambe le impostazioni e degli strumenti coinvolti, presentiamo ampi risultati sperimentali e studi di ablazione che dimostrano il valore delle tecniche proposte in questa ricerca.

This work investigates the entanglement between Continual Learning (CL) and Transfer Learning (TL). In particular, we shed light on the widespread application of network pretraining, highlighting that it is itself subject to catastrophic forgetting. Unfortunately, this issue leads to the under-exploitation of knowledge transfer during later tasks. On this ground, we propose Transfer without Forgetting (TwF), a hybrid Continual Transfer Learning approach building upon a fixed pretrained sibling network, which continuously propagates the knowledge inherent in the source domain through a layer-wise loss term. Our experiments indicate that TwF steadily outperforms other CL methods across a variety of settings, averaging a 4.81% gain in Class-Incremental accuracy over a variety of datasets and different buffer sizes.