Mapping the spatial patterns of synaptic connectivity is essential for understanding neural circuit function. Two-photon optogenetics combined with whole-cell patch clamp enable synaptic circuit mapping in mammalian cortex. We consider an experimental set-up in which the experimenter patches a single target cell and uses two-photon optogenetics to stimulate surrounding tissue in a 3-dimensional grid while measuring postsynaptic currents (PSCs). This yields a high resolution spatial map of putative synaptic connections to a single neuron. However, these connectivity maps are noisy due to biological variability and measurement noise. Typical image denoising methods ignore the probabilistic nature of upstream spikes and thus do not capture the full response distribution. We propose a deep learning framework to overcome this limitation. Our method uses a single Convolutional Neural Network (CNN) to predict the distribution of PSC amplitudes at each stimulated location-power combination. We train the network using a large dataset of simulated experiments built from voxelized electron microscopy (EM) cell reconstructions, in which we have access to the ground truth cell locations, morphologies, and synaptic weights. Our network is trained to output a discretized representation of the response cumulative distribution function (CDF) at each voxel. The CDF, unlike the mean, describes both response reliability and putative connection strength. On data from mouse V1, we show that our method outputs more interpretable response maps than similar denoising methods. Unlike iterative methods, the proposed approach requires only a single forward pass through a neural network, meaning that it can be deployed in real time to guide experiments in progress and scales easily to large tissue samples. Our simulator includes a generative model of upstream spikes and is available as an open-source software library. These tools will allow experimenters to characterize the strength of synaptic connections across space at a level of granularity impossible with prior techniques.