Search for: All records

Deep learning based PET image reconstruction methods have achieved promising results recently. However, most of these methods follow a supervised learning paradigm, which rely heavily on the availability of high-quality training labels. In particular, the long scanning time required and high radiation exposure associated with PET scans make obtaining these labels impractical. In this paper, we propose a dual-domain unsupervised PET image reconstruction method based on learned descent algorithm, which reconstructs high-quality PET images from sinograms without the need for image labels. Specifically, we unroll the proximal gradient method with a learnable norm for PET image reconstruction problem. The training is unsupervised, using measurement domain loss based on deep image prior as well as image domain loss based on rotation equivariance property. The experimental results demonstrate the superior performance of proposed method compared with maximum-likelihood expectation-maximization (MLEM), total-variation regularized EM (EM-TV) and deep image prior based method (DIP). 

Inspired by humans’ exceptional ability to master arithmetic and generalize to new problems, we present a new dataset, Handwritten arithmetic with INTegers (HINT), to examine machines’ capability of learning generalizable concepts at three levels: perception, syntax, and semantics. In HINT, machines are tasked with learning how concepts are perceived from raw signals such as images (i.e., perception), how multiple concepts are structurally combined to form a valid expression (i.e., syntax), and how concepts are realized to afford various reasoning tasks (i.e., semantics), all in a weakly supervised manner. Focusing on systematic generalization, we carefully design a five-fold test set to evaluate both the interpolation and the extrapolation of learned concepts w.r.t. the three levels. Further, we design a few-shot learning split to determine whether or not models can rapidly learn new concepts and generalize them to more complex scenarios. To comprehend existing models’ limitations, we undertake extensive experiments with various sequence-to-sequence models, including RNNs, Transformers, and GPT-3 (with the chain of thought prompting). The results indicate that current models struggle to extrapolate to long-range syntactic dependency and semantics. Models exhibit a considerable gap toward human-level generalization when evaluated with new concepts in a few-shot setting. Moreover, we discover that it is infeasible to solve HINT by merely scaling up the dataset and the model size; this strategy contributes little to the extrapolation of syntax and semantics. Finally, in zero-shot GPT-3 experiments, the chain of thought prompting exhibits impressive results and significantly boosts the test accuracy. We believe the HINT dataset and the experimental findings are of great interest to the learning community on systematic generalization. 

The physical mechanisms that link the termination of star formation in quiescent galaxies and the evolution of their baryonic components, stars, and the interstellar medium (ISM; dust, gas, and metals) are poorly constrained beyond the local Universe. In this work, we characterise the evolution of the dust content in 545 quiescent galaxies observed at 0.1 < z < 0.6 as part of the hCOSMOS spectroscopic redshift survey. This is, to date, the largest sample of quiescent galaxies at intermediate redshifts for which the dust, stellar, and metal abundances are consistently estimated. We analyse how the crucial markers of a galaxy dust life cycle, such as specific dust mass (M_dust/M_⋆), evolve with different physical parameters, namely gas-phase metallicity (Z_gas), time since quenching (t_quench), stellar mass (M_⋆), and stellar population age. We find morphology to be an important factor in the large scatter inM_dust/M_⋆(∼2 orders of magnitude). Quiescent spirals exhibit strong evolutionary trends of specific dust mass withM_⋆, stellar age, and galaxy size, in contrast to the little to no evolution experienced by ellipticals. When transitioning from solar to super-solar metallicities (8.7 ≲ 12 + log(O/H)≲9.1), quiescent spirals undergo a reversal inM_dust/M_⋆, indicative of a change in dust production efficiency. By modelling the star formation histories of our objects, we unveil a broad dynamical range of post-quenching timescales (60 Myr < t_quench < 3.2 Gyr). We show thatM_dust/M_⋆is highest in recently quenched systems (t_quench < 500 Myr), but its further evolution is non-monotonic, as a consequence of different pathways for dust formation, growth, or removal on various timescales. Our data are best described by simulations that include dust growth in the ISM. While this process is prevalent in the majority of galaxies, for ∼15% of objects we find evidence of additional dust content acquired externally, most likely via minor mergers. Altogether, our results strongly suggest that prolonged dust production on a timescale of 0.5 − 1 Gyr since quenching may be common in dusty quiescent galaxies at intermediate redshifts, even if their gas reservoirs are heavily exhausted (i.e. cold gas fraction < 1 − 5%). 

Abstract Cyclic degradation in flexible electronic inks remains a key challenge while their deployment in life critical applications is ongoing. The origin of electrical degradation of a screen-printed stretchable conductive ink with silver flakes embedded in a polyurethane binder is investigated under uniaxial monotonic and cyclic stretching, using in-situ confocal microscopy and scanning electron microscopy experiments, for varying ink thickness (1, 2, and 3 layers, each layer around 8–10 μ m) and trace width (0.5, 1, and 2 mm). Cracks form under monotonic stretching, and the evolution of crack pattern (density, length and width) with applied strain is affected by ink thickness such that the 3-layer ink exhibits larger normalized resistance but slightly lower resistance than the 1-layer ink up to strains of 125%. For cyclic stretching, the crack density and length do not evolve with cycling. However, the cracks widen and deepen, leading to an increase in resistance with cycling. There exists a strong correlation between fatigue life, i.e. the number of cycles until a normalized resistance of 100 is reached, and the strain amplitude. The normalized resistance increase rate with respect to cycling is also found to scale with strain amplitude. The rate of change in resistance with cycling decreases with ink thickness and trace width. For practical applications, thicker ( ⩾ 25 μ m) and wider (⩾2 mm) inks should be used to lower resistance increases with repeated deformation.